How does one tell whether a word has one, several or many senses?

1 Mr Wiggins is certainly correct in supposing that the answer to the title question should be based on an account of what it is for a word to have a certain meaning, and the associated account of how specifications of word-meaning are to be tested. If we take a Lockean view according to which having a meaning is a matter of being regularly used as the sign of a certain (Lockean) idea, then the job of showing that ‘necessary’ has several different meanings is quite a different affair from what it would be if we accept the view that for a word to have a certain meaning is for utterances of that word to stand in certain causal relations with physical stimuli and/or overt responses. For my part I share with Mr Wiggins, and many other theorists, the conviction that for a word to have a certain meaning is for it to make a certain contribution to some appropriate semantic property of sentences in which it occurs, a property that we may dub ‘sentence-meaning’. I say ‘dub’ because I doubt that there is enough talk about the meaning of sentences (unlike the situation vis-a-vis words), to yield any substantial pre-theoretical concept properly so called. The kind of sentence property that I believe both Mr Wiggins and I (along with others) have in mind is most perspicuously indicated in ordinary language by some such term as ‘what the sentence can be used to say’ (in a sense of say in which ‘what he said’ is not synonymous with ‘what sentence he uttered’), or more barbarously, the sentence’s ‘saying potential’. Mr Wiggins wishes to explicate this notion in terms of the conditions under which an indicative sentence can be used to make a true statement (or, in his lingo, the conditions under which an utterance of the sentence will be factually licensed). He correctly anticipates that some will find this approach intolerably restricted, and I count myself among that number. I think it not too vaulting an ambition to search for a more general account of ‘ saying potentials ’ of sentences, such that the potentiality of sentences of a certain sub-class for being used to make statements would be a special case of this more general notion. I have elsewhere provided the beginnings of such an account, using the term ‘illocutionary act potential’.¹ (This is not Austin’s concept of an illocutionary act, whatever that is, though I did filch the term from him.) Roughly, for a sentence to have a certain illocutionary act potential is for it to be subject to a rule that enjoins members of the language community from uttering the sentence, in certain kinds of contexts, unless certain specified conditions hold.

Thus the central illocutionary act potential of ‘Please pass the salt’ would be specified by making it explicit that a rule is in force in the English language community that could be roughly formulated as follows:

One is not to utter ‘Please pass the salt’ in a normal ² context unless the following conditions hold:

1. Someone, H, is being addressed.

2. It is possible for H to pass the speaker some salt.

3. The speaker has some interest in getting H to pass him some salt.³

For the special case where the illocutionary act, a potential for which is in question, is something statemental or assertive, something that is straightforwardly true or false, the conditions imbedded in the rule would coincide with the truth conditions for the statement. Thus the most basic illocutionary act potential for ‘My uncle sold his wireless’ could be specified by making it explicit that the following rule is in force in the English language community:

One is not to utter ‘ My uncle sold his wireless ’ in a normal context unless the following conditions hold:

1. Some particular uncle, x, of the speaker is contextually indicated.

2. At some time prior to the time of utterance, x sold a wireless that had belonged to him.

Thus the filling for the specification of illocutionary act potential for this sentence would coincide with what Wiggins calls conditions for the utterance being factually licensed. Any doubt as to whether a certain set of conditions, e.g. those listed above, do constitute the truth conditions for the statement (or conditions for an utterance of the sentence being factually licensed), would equally be a doubt as to whether the sentence is subject to a rule of the above sort that requires just those conditions to hold.

Instead of continuing the exposition of my views on sentence-meaning, I shall address myself in this paper to problems that have specifically to do with distinguishing senses of words. Let us take it as agreed between Wiggins and myself that we have some appropriate sense of ‘sentence-meaning’, and that we are thinking of a meaning of a word as a constant contribution it makes to the meaning of any sentence in which it occurs with that meaning. Then how are we to conceive the job of determining how many senses a word has? Again I find myself in sympathy with the general thrust of Wiggins’ account. A word is properly assigned as many senses as is necessary to account for the facts about the meanings of sentences in which it occurs. I should like to put more stress than Wiggins does on the ineradicably systemic character of the evaluation of particular semantic hypotheses, given this approach to meaning. (Though I have no reason to think he would disagree with what I am about to say.) If dictionary entries are to be evaluated in terms of the ‘readings’ or ‘ interpretations ’ of sentences they yield, any given entry will have to be evaluated in conjunction with a number of other semantic hypotheses. Suppose that I have two p oposed entries for ‘run’ (as a transitive verb), very roughly as follows: (a) operate, (b) force, together in each case with what Wiggins calls ‘directives’ and what Katz calls ‘selection restrictions’. ⁴ Obviously we can’t use these entries by themselves to derive any readings for any sentences, e.g., ‘How long have you been running the engine? ’ or ‘You are going to run me into debt’.

To derive readings for these sentences we need not only entries for ‘run’, but also entries for the other words, specifications of the syntactical structure of the sentences and constituent complex expressions, and what Katz calls ‘projection rules’,⁵ which for a given kind of complex expression tell us how to go from facts like the above to a reading for the whole expression. Thus on this approach dictionary entries cannot be tested in isolation, one-by-one, any more than can any other constituents of the semantic description of a language. Just as a particular general hypothesis in science can be empirically tested only in the context of some system of principles, so with particular semantic and syntactical hypotheses about a particular language.

This means that our title problem merges into the very large methodological problem: How can we formulate and evaluate a semantic description of a language? For the above considerations show that my justification for supposing that, e.g., ‘ run ’ has at least two meanings, operate and force, can be no greater than my justification for supposing that an adequate lexicon for English would include two such entries for ‘ run ’. The most fundamental justification of this latter claim would be, of course, the demonstration that a given lexicon containing theses entries is an adequate one, and, as indicated above, we could not show that without showing that an adequate semantic description of the language could be constructed using this lexicon.⁶ Thus an ideally thorough treatment of this particular methodological problem would include consideration of all the methodological problems involved in the formulation and evaluation of the semantic description of a language and the components thereof. This includes such problems as:

1. What form should dictionary entries take?

2. What form should the specifications of sentence-meaning take?

3. What facts about sentence-meaning can be used as data for the testing of semantic systems, and how can these facts be gleaned from the behavior of language users?

4. All the methodological problems involved in developing the syntactical description of a language.

These problems do not come piecemeal. Language itself is so systematic that the investigation of language fails to be equally systematic only at its peril.

2 Rather than emit obiter dicta on such large issues as these, I shall turn to the consideration of a more modest methodological problem, one small enough to be illuminated in the time at my disposal. Suppose that we find ourselves required to decide whether a given word is used in the same or different senses in two contexts,⁷ but we do not have the time, resources, or ingenuity to consider this problem in the light of some proposal for a complete semantic description of the language. This is a situation in which lexicographers and philosophers often find themselves. For the traditional lexicographer such problems come up in the course of deciding on a set of entries for a word in a dictionary of the usual sort, one with much more modest pretensions than the ideal lexicon about which we have been talking. For the philosopher, on the other hand, they most usually arise in the course of giving analyses of the concepts expressed by certain terms.

If a philosopher is concerned to analyze the concept expressed by ‘remember’ over a certain stretch of contexts, he will have to determine how many senses the word has over that stretch, so that he will know how many concepts to analyze. Again, if a philosopher A claims that ‘true’ as it occurs in

(1) What you say is true.

means corresponds to the facts, and then B attacks this by pointing out that one can be a true friend without corresponding to any facts, then if A is to defend his analysis he will have to either show that B’s claim is mistaken, or show that in

(2) He is a true friend.

‘true’ is used in a sense different from the one in which he is interested. My relatively modest methodological question is: what devices, short of a consideration of schemes for a complete description of the language, can we deploy for the resolution of such questions? And what difficulties will we encounter in this enterprise?

It may be useful to sharpen our intuitions and initially formulate our principles in the context of some pedestrian examples far removed from difficult and philosophically exciting terms like ‘true’ and ‘remember’. For this purpose, consider the occurrences of the humble term ‘run’ in

(3) Harold Stassen is still running.

(4) The boundary ran from this tree to that tree.

(5) John ran from this tree to that tree.

(6) The engine is still running.

(7) Is the vacuum cleaner still running?

We may pose as our initial sample problems: how can we show that ‘run’ has at least two senses in (3), but only one sense in (4)? And how can we show that ‘run’ has different senses in (4) and (5), but the same sense in (6) and (7)?

If these are to be precise questions, we shall have to make explicit what resources we are allowing our inquirer. I suggest that we regard him as capable of making reliable judgments of (at least approximate) sameness and difference of sentence meaning. Thus he can determine that (6) and (7) have different meanings, that (4) and

(8) The boundary extended from this tree to that tree.

have the same meaning, and that (3) has at least two different meanings. Moreover we shall regard him as capable of determining what entailment relations hold between sentences, e.g., that (6) entails ‘ The engine is in working condition ’. That is, we are going to trust his ‘ intuitions ’ about the semantic relations of sentences taken as (semantically) unanalyzed units, even though, lacking a fine grained semantic description of the language, or even a proposal for such, he is unable at this stage to analyze sentence-meanings into the components contributed by the words that make up the sentence. In granting him even these relatively crude capacities we are making some large, and recently highly controversial, assumptions about the possibility of drawing a line between what one means by a sentence and what happens to be true of the subject matter, between the analytic and the synthetic, and so on. But for purposes of this paper we can do no more than note these assumptions in passing.

Proceeding, armed with these tools, to our sample problems, the first thing to note is that the tools do not suffice in and of themselves for the resolution of the problems. It does not follow just from the fact that (4) and (5) have different meanings that ‘run’ is used in a different sense in them. To establish that we have to rule out the possibility that other features of this sentence pairing, e.g., the interchange of ‘The boundary’ and ‘John ’, are solely responsible for the difference in sentence-meaning. After all, (6) and (7) differ in meaning too; but here we are not inclined to attribute this difference, even in part, to a difference in the sense of ‘run’. Again, admitting that (3) can be used to say two quite different things, why attribute this difference to differences in senses of ‘run’? The mere fact that there are two different meanings of this sentence is compatible with a number of alternative hypotheses, e.g., that one of the other words, e.g., ‘still’, has two different senses in this context, or that the sentence can be grammatically construed in two different ways. And yet we are strongly inclined to attribute the plurality of sentence-meanings to a plurality in the senses of ‘run’. How can these inclinations be justified?

If we had adequate analyses of the meanings of each of these sentences, we could read off from those analyses what each sentence-constituent contributes to the whole. Hence we could say what the contribution of ‘run’ is to the meaning of (4) and to the meaning of (5), thereby determining whether it is the same contribution in each case. But this is just what we do not have, short of an adequate systematic semantics of English. What we are looking for is the closest approximation to this procedure that is possible within the specified limitations. A technique that looks promising is that of partial substitution. Although we cannot say what ‘ run ’ contributes to the meaning of (4) and of (5), we can try various substitutions for ‘run’ in each of the two sentences, noting which ones do and which ones do not preserve sentence meaning. Thus we can preserve (approximately) the meaning of (4) while substituting ‘extend’ for ‘run’, and we can preserve (approximately) the meaning of (5) while substituting ‘locomote springily’⁸ for ‘run’, but opposite substitutions will produce a marked change in (or destroy) sentence-meaning. Similarly, if we consider two contexts in which (3) would be used to say different things, in one of these contexts sentence-meaning can be preserved by a substitution of ‘actively seek public office ’ for ‘ run ’, in the other context sentence-meaning can be preserved by a substitution of ‘ locomote springily ’ for ‘ run ’, but again not vice versa. These results support the claim that at least part of the difference in sentence-meaning is due to differences in the senses with which ‘run’ is used. What this technique gives us is an indirect, symptomatic approach as a substitute for an unavailable direct account of the underlying structure. This technique is related to the ideal demonstration as a symptomatic diagnosis of a disease is related to a pathological description. The fact that non-convertible replacements for ‘run’ in (4) and (5) will preserve sentence meaning is taken as a symptom of a change of sense in ‘run’ between (4) and (5), just as an increase in body temperature is taken as a symptom of an increase in infection somewhere in the body. Let us consider the assumptions we make in performing this symptomatic inference. I will begin with the same-sentence case, since the other case involves all the assumptions present there, in addition to some others.

Let us consider two situations, S₁ and S₂, in which (3) is being used to say two different things. In S₁ we can say the same thing by

(9) Harold Stassen is still locomoting springily.

In S₂ we can say the same thing by

(10) Harold Stassen is still actively seeking public office

but not vice versa. This would seem to show conclusively that in S₁ ‘run’ is being

used to mean the same thing as ‘locomote springily’ but not the same thing as ‘actively seek public office’ and in S₂ vice versa. Since ‘locomote springily’ and ‘actively seek public office’ are not interchangeable in this sentential context, they do not mean the same thing in this context. Therefore, since ‘run’ is synonymous with the one in S₁ and with the other in S₂, it must have different meanings in S₁ and S₂.

However, without falling into the last extremities of Cartesian scepticism we can note the following possibility, which, if realized, would throw off this inference. Suppose that some other word or words in the sentence frame ‘Harold Stassen is still — ’ shifts its meaning in the course of one of these substitutions, e.g., the substitution of ‘actively seeking public office’ for ‘running’. In that case the fact that in S2 (Io)has the same meaning as (3) would not show that in S2 ‘run’ has the same meaning as ‘ actively seek public office ’. It would show the opposite. For if some part of the frame changes its meaning, and the substitute for ‘run’ is a synonym, then the resulting sentence would have a different meaning from (3). Under those conditions, (10) would have the same meaning as (3) only if ‘ actively seeking public office ’ were to differ in meaning from ‘ running ’ in such a way as to neutralize the other differences in the sentences. By the same argument we can show that if some part of the frame were to shift its meaning through the substitution, the fact that (3) (in S2) and (9) differ in meaning does not show that ‘run’ and ‘locomote springily’ differ in meaning in this context. For even if they have the same meaning in this context, (3) and (9) would differ in meaning because of the semantic shift elsewhere in the sentence. These considerations show that in using the substitution test we are assuming that the rest of the sentence holds fast semantically. If we had to justify such an assumption we would have to carry out substitutions with respect to each of the other constitutents of the sentence. We would have to show, e.g., that ‘still’ can be replaced by ‘yet’, salva meaning in both (3) and (9). But this argument would be subject to analogous assumptions concerning the semantic fixity of the rest of these sentential contexts, as well as the assumptions that ‘yet’ has the same meaning in the two sentences. Hence the attempt to justify every such asumption one-by-one would lead to an infinite regress. It would seem that at some point we are forced to invoke a principle of simplicity, according to which terms are held to retain the same meaning over two contexts unless we are forced to recognize a difference.

Consider now the case where we make non-convertible substitutions in different sentences, e.g. (4) and (5). (5) has the same meaning as

(11) John locomoted springily from this tree to that tree.

but not the same meaning as

(12) John extended from this tree to that tree.

whereas in (4) ‘ ran ’ is replaceable by ‘ extended ’ but not by ‘locomoted springily ’. From this we infer that ‘ran’ has different meanings in (4) and (5). Here the reasoning is subject to the same assumption as in the same-sentence case, viz., that in each substitution the rest of the sentence remains semantically fixed; but there is an additional problem that stems from the difference, in sentence-contexts. For suppose that ‘extended’ shifts its meaning from (8) to (12). In that case the fact that (4) preserves the same meaning under a substitution of ‘extended’ for ‘ran’ while (5) does not, does not show that ‘ran’ does not have the same meaning in (4) and (5). For even if ‘ ran ’ has the same meaning in both contexts, we would not expect that meaning to be preserved in both contexts under a substitution of ‘ extended ’ for ‘ ran ’, if ‘ extended ’ has a different meaning in the two contexts. We would expect just the reverse. Thus the substitution test employed here assumes that our substituends, ‘locomoted springily’ and ‘extended’, do not have different meanings in the two contexts. And again any attempt to justify this assumption would lead us to make further analogous assumptions.

I uncover these assumptions not in order to show that we are never justified in concluding that a word has two different senses in two contexts, a conclusion I would not embrace, but rather to re-emphasize the systemic character of investigations in this area, as in other areas of language. If we had a workable semantic description of the whole language, each component would receive its justification from its presence in the system, which in turn would be justified by the fact that it would do what we expect a semantic description of a language to do.⁹ But so long as we are confined to piecemeal inquiries we will correspondingly be forced, at any stage, to rely on assumptions of sameness and difference of meaning that have received no justification. While this is our condition we are undoubtedly well justified here, as elsewhere, in proceeding on the basis of principles of simplicity like the one cited two paragraphs back, and in these terms we are often justified in drawing conclusions as to sameness of difference of sense. But ideally such conclusions are provisional, pending a systematic development of a total semantics of the language.

It is not in every case of a suspected difference in meaning that we are able to carry out substitutions just for the word in question. Sometimes the sentence meaning cannot be so neatly dissected, and we have to get along with still more indirect indications. Consider

(13) I ran him a close second.

and

(14) He always runs everything together.

I daresay we will not find any substitution for ‘ run ’ in either of these sentences which, while leaving the rest of the sentence unchanged, will preserve sentence meaning. We can find near-equivalents of the two sentences as wholes. Thus (13) has about the same meaning as

(15) I placed second, close behind him.

and (14) has about the same meaning as

(16) He fails to distinguish things sufficiently.

(15) and (16) are so different from each other that it encourages us to say that ‘run’ must have some different meaning in (13) and (14). Moreover the fact that (15) is concerned with the speaker’s rank in the results of some contest, while (16) is concerned with failing to make distinctions, encourages us to say that in (13) ‘run’ means something like be ranked as, while in (14) ‘ run ’ means something like confuse. But only something like. After all, what we have in (15) and (16) are paraphrases of whole sentences in which several constituents are replaced and the structure changed. Hence this technique fails to pinpoint the semantic contribution of ‘run’.

The same disability attaches to the demonstration of mutually non-substitutable entailments. (13) but not (14) entails that the speaker was in some contest, while (14) but not (13) entails that the person referred to has been talking about several distinguishable topics. But again it is the whole sentence that has this implication in each case, and there remains a question as to what this shows us about the specific semantic contribution of ‘run’.

In these cases our conclusions do not have even the kind of provisional validity enjoyed by those based on the word-substitution test. Here our grounds are shakier because less explicit. In the case of the former we could identify assumptions such that if these assumptions were correct the conclusions would be established. Furthermore the assumptions were of the same type as the conclusion in question, and so any one of them could itself be tested, subject, of course, to the same necessity for dependence on other like assumptions. Nothing like that is possible here. Since we have no resources, analogous to the word-substitution test, for prying the sentence apart semantically, we are unable to identify any testable assumptions on which our conclusion depends.¹⁰ Such basis as we have is intuitive rather than discursive. It seems to us, as we mouth the sentences, that the word ‘run’ is ‘doing something different’ in the two cases. But of course such impressions are notoriously fallible. That is not to say that they are worthless. We frequently have to make do with this sort of thing in an undeveloped stage of a discipline, and the impressions of sensitive trained observers are by no means to be taken lightly. Nevertheless it is salutary to realize just what status our conclusions have.

3 I should now like to apply the results to a consideration of some cases in which philosophers find themselves called on to decide questions of sameness or difference of meaning. In many such cases the philosopher is able to make use of the pinpointed substitution test. Thus we can show that in (1) but not (2) ‘true ’ can be replaced by ‘correct’, while in (2), but not in (1), ‘true’ can be replaced by ‘ real ’. Again, a philosopher doing philosophical psychology may want to distinguish the senses of ‘ want ’ in the most common uses of

(17) I want an ice cream soda.

and

(18) That child wants a good spanking.

This he can do by pointing out that in (17) but not (18) ‘want’ can be replaced by ‘have a desire for’, while in {18) but not (17) ‘want’ can be replaced by ‘need’. These conclusions will have the status we earlier saw such conclusions to have. That is, they can be considered justified, given certain plausible assumptions that can either be accepted on a simplicity principle or investigated in the same way as the conclusions in question.

Wiggins’ contentions about ‘good’ employs this technique. I take him to be claiming that

(19) She has good legs.

has several meanings such that in one and only one of these meanings ‘ good ’ can be replaced by ‘ beautiful ’ salva sentence-meaning, in one and only one meaning of the sentence ‘good’ can be replaced by ‘healthy’ and so on. Let me just note in passing that the only serious opposition to Wiggins’ thesis will come at the level of sentence-meaning. One who wishes to maintain the univocity of ‘good’ will (be well advised to) claim that (19) is used with the same meaning (is used to make the same assertion) in all the contexts of which Wiggins is thinking, and that the differences to which he alludes are differences in the considerations that are relevant to the evaluation of this one and the same assertion in one or another context. In other words, the defender of univocity would challenge Wiggins’ claim that (19) sometimes means the same as ‘She has beautiful legs ’, sometimes the same as ‘She has healthy legs ’, etc. He would maintain that each of those latter sentences means more than (19) ever means, for it adds to the claim made by (19), a claim about the proper criteria of evaluation for that context. Thus the main issue here has to do with the relation of meaning and criteria for evaluative terms (or sentences). I do think that if Wiggins’ claims on the sentence-meaning level are granted, there can be no serious doubt that he is correct about the word ‘good ’.

However the philosopher is not always in a position to carry out the word-substitution test. Suppose that a philosopher has proposed an analysis of ‘see’ such that one of the defining conditions consists of the actual existence of the object of sight in the physical environment of the perceiver. Now he is confronted with locutions like

(20) I see Mt Rainier before my mind’s eye.

where there is obviously no implication that Mt Rainier is actually in the physical environment. He replies, of course, that this is a different sense; he means to be elucidating the concept of seeing as the exercise of a sense organ, not the concept of the exercise of visual imagination. But how can he show that the word ‘see’ is used in different senses in (20) and

(21) The clouds have lifted; I now see Mt Rainier.

He can, of course, find equivalents to (20) of a sort not available for (21). Thus (20) is roughly equivalent to ‘I have a mental image of Mt Rainier’ and to ‘I am visualizing Mt Rainier’, while (21) is paraphrasable in no such ways. But does this show that ‘see’ is used in different senses in (20) and (21)? A partisan of the single sense view will (be well advised to) concede that these results show that ‘see x in the mind’s eye’ has a different meaning from ‘see x ’. For the former phrase, but not the latter, can be replaced by ‘have a mental image of x’ or ‘am visualizing x ’. But this is not at issue. The question is as to whether the word ‘see’ itself makes a different contribution to the meaning of the two phrases. One who maintains that it does not can hold that the difference in meaning in (20) and (21) comes just from the presence of ‘in the mind’s eye’ in (20) and not in (21). And unless we can pry apart these phrases by making substitutions just for ‘see ’, we will have no argument against him.

Again suppose that H. P. Grice is attempting to separate out and analyze a specially ‘ semantic ’ sense of ‘ mean ’ as used of speakers, a sense of ‘ mean ’ in ‘ What S meant (by what he said) was —’, such that from a statement of this form we can derive a specification of the sense in which he was using whatever sentence he uttered, or a specification of what the sentence meant as he was using it on that occasion.¹¹ In order to focus on this we shall have to filter out other senses of ‘mean’ as used of speakers, e.g., in

(22) What do you mean?

together with answers thereto. Sometimes (22) is a request for a justification of what was said (‘I’m not going to the party’, ‘What do you mean, you’re not going’?). Sometimes (22) is a request to be more specific (Doctor: ‘You’ll be out of the hospital soon.’ ‘What do you mean, “soon”?’).

We shall want to distinguish these senses of (22) and correlated senses of ‘I mean —’ from the case in which (22) is a request for a more intelligible paraphrase of the sentence uttered. But can we show that ‘mean’ has different meanings, in (22), in cases of these three sorts? Let us agree that the equivalents of the whole sentence are different in the three cases. In one it is ‘What is your justification for saying that? ’, in another it is ‘Be more specific ’, in the third, ‘Give me a paraphrase of your sentence that I can understand ’. But how do we show that these differences in sentence-meaning are due to differences in the sense of ‘ mean ’ rather than, or as well as, differences in the meaning of ‘ what ’ or ‘do’, or differences in grammatical structure? It certainly seems intuitively plausible to suppose that differences in the sense of ‘mean’ must be at least partially responsible, but how to show this?

Again, suppose that in the course of trying to understand emotion-concepts, I feel the need to distinguish ‘occurrent’ from ‘dispositional’ uses of emotion terms. That is, I want to show, e.g., a difference in the senses of ‘afraid’ in

(23) I have always been afraid of snakes.

and

(24) When he started toward us I became very afraid of him.

The search for non-convertible replacements for ‘afraid’ alone is doomed to failure because any substitution that will work in one, e.g., ‘frightened’, will work in the other as well. If ‘ afraid ’ has different senses in these contexts, so does ‘ frightened ’. We may then be led to pointing out differences in entailment patterns. Thus (23) entails ‘Usually when I see snakes I get frightened’, but there is no comparable entailment for (24), e.g., ‘ Usually when he starts toward us, I get frightened’. But of course this is a difference between (23) as a whole and (24) as a whole. We still have not shown that this difference is due, at least in part, to a difference in senses of ‘ afraid ’.

In these three cases our conclusions have the purely intuitive character we have seen earlier to attach to such conclusions. Again this is not to claim that the conclusions should be abandoned, but only to point out where we are at this stage of the game.

I have been surveying cases in which the philosopher is in a relatively good position to establish difference of meaning, and less favorable cases in which, nevertheless, it seems plausible to suppose that there is a difference of meaning which would be established with more adequate devices. Now I want to call attention to some cases in which philosophers make blatantly unjustified claims of multivocality through neglect of some of the points set out above.

Often when I feel inclined to cavil at multivocality claims the controversy is properly located at the sentence-meaning level. Thus B. F. McGuiness maintains that ‘want’ must be used in a different sense when talking of conscious and of un¬ conscious wants.¹² He bases this claim on the claim that ‘P wants S’ entails ‘P knows that he wants S’ when we are using ‘want’ in an ordinary sense of the term, but not when we are talking about repressed wants. (Hence a sentence of the form ‘P wants S’ means something different in these two contexts.)

If he is right about this difference in sentence-meaning, his multivocality claim is surely justified, but I see no reason to accept the entailment claim. Again Norman Malcolm notoriously maintains that ‘He had a dream last night’ has a meaning different from its ordinary meaning if it is based on REM evidence rather than on the subject’s reports on awakening.¹³ For the sentence will be logically related to different ‘criterial’ statements in the two cases. Again if this difference of sentence-meaning can be established, it will be at least very plausible to suppose that the word ‘dream’ has different meaning in the two contexts. But again I do not feel constrained to recognize that what I am prepared to take as decisive evidence enters into meaning in the way Malcolm supposes.

From the point of view of this paper the more interesting cases of philosophical folly are those in which the putative differences in sentence-meaning are uncontroversial, but where word-multivocality is inferred from this without heeding the cautions insisted on earlier. The most flagrant cases are those in which a multiplicity of senses is ascribed to some term on the ground that different sentences in which it occurs are tested or verified or established differently. Thus ‘ there is ’ is said to have different meanings in

(25) There is a fireplace in my study,

and

(26) There is a prime number between 6 and 10.

on the grounds that the justifications of these two statements are widely different. There is no doubt that the justifications are widely different and that {25) and (26) have different meanings (not that we needed an appeal to verifiability to show that). However this is radically insufficient to show that ‘there is’ has different meanings in the two sentences. It would seem that the difference in meaning between ‘ fireplace in my study’ and ‘prime number between 6 and 10’ is quite sufficient to yield the difference in meaning between (25) and (26) that is reflected in the different justifications. Analogous remarks are to be made about the similar argument that ‘true’ has different meanings in

(27) It is true that there is a fireplace in my study,

and

(28) It is true that there is a prime number between 6 and 10.

on the grounds that (27) and {28) are tested in radically different ways. Once more the difference in what follows ‘ It is true that’ in the two cases is quite sufficient to account for that difference.

Again philosophers will argue for multivocality from differences in patterns of entailment, heedless of the possibility that these differences may be explained by differences in other constitutents of the sentence. Thus ‘know’ is said to have different senses in

(29) I know that I feel disturbed,

and

(30) I know that my car is in the garage.

on the grounds that (29) is entailed by ‘ I feel disturbed’,¹⁴ while (30) is not entailed by ‘ My car is in the garage ’. But it still remains to be shown that this difference in entailment patterns reflects a difference in the meaning of ‘ know ’, rather than just a difference between ‘ I feel disturbed ’ and ‘ My car is in the garage ’. Again, it is common for philosophers to hold that even when we restrict ourselves to specifications of linguistic meaning ‘ mean ’ has a different meaning depending on whether we are talking of a ‘ categorematic ’ word like ‘ thermometer ’ or a ‘ syncatetorematic ’ word like ‘if’.

This claim is rarely given any justification other than a question-begging one to the effect that a favored analysis of the concept of meaning (usually of a referential sort) applies to the former but not to the latter. But one might try to justify the claim by pointing out that ‘ “ Thermometer ” means instrument for measuring temperature ’ entails that ‘ thermometer ’ denotes instruments for measuring temperature, whereas ‘“If” means provided that’ does not entail that ‘if’ denotes ‘provided that’. However, even if we grant this entailment claim, it will not follow that ‘mean’ has different senses in the two contexts. For the difference in entailments may be adequately accounted for by the differences between ‘thermometer’ and ‘instrument for measuring temperature’ on the one hand and ‘ if ’ and ‘ provided that ’ on the other. In other words, it may be that these expressions just have different kinds of meanings, not meanings in different senses of ‘mean’.

4 We have seen that the attempt to decide questions of multivocality piecemeal is beset with serious difficulties. In the most favorable cases we have to depend on assumptions that receive such justification as they have from considerations of simplicity. In less favorable cases our conclusions rest on unadulterated intuition, pending more systematic constructions. Now it may be that these difficulties are not basic problems for the philosopher just because, contrary to first impressions, the philosopher, unlike the lexicographer, does not have to settle questions of word-multivocality. It is true that most philosophers who have thought of conceptual analysis as a linguistic enterprise have thought of it as essentially concerned with the meanings of words. Thus much meta-ethics has been concentrated on questions about the meaning(s) of ‘ good ’, there has been much talk in epistemology about the meaning(s) of ‘know’ and so on. But it may not be necessary for the analytical philosopher to couch his problems in this way. If one is trying to understand value-judgments, or moral judgments, it is not essential for the accomplishment of that purpose to provide a set of dictionary entries for ‘ good ’ that would be satisfactory as a part of an adequate semantic description of English. To meet that requirement the entries would have to be such as to yield, together with the appropriate other parts of the description, acceptable interpretations of any sentence, declarative or otherwise, in which ‘good’ occurs. However it seems that one could arrive at a philosophically illuminating understanding of value-judgments just by giving patterns of interpretation for sentence-types, like ‘x is a good ψ without attempting to spell out what is contributed to the interpretation of the whole sentence by each of its meaningful components. (Many philosophers would prefer to call such results ‘contextual definitions’ of ‘good’, for various contexts). Again if I am interested in understanding linguistic meaning, then I am interested in giving interpretations of sentences of the form ‘x means y’, where x is a variable ranging over linguistic expressions, sentences of the form ‘ P knows what x means ’, etc. It is not essential for this purpose that I determine whether or not ‘ mean ’ has the same sense in these contexts as it does in various contexts where it is used in conjunction with something other than designations of linguistic expressions, e.g., ‘ I mean Susie ’, or ‘ That look on his face means trouble’. Here too we can give interpretation of the sentence-types with which we are concerned without going into their fine semantic structure. Again, philosophers who make claims about ‘ there is ’ and ‘ true’ of the sort we criticized a few pages back may not have to make such claims in order to accomplish their main purposes. It may be that they too can restrict themselves to the sentence-level, e.g., by just pointing out the differences in what we are saying in uttering sentences like (25) and (26), without having to trace these differences to differences in the meanings of ‘ there is ’ in these sentences.

I am not claiming that it is never important for the philosopher, qua philosopher, to provide adequate dictionary entries, or to establish word-multivocality. The boundaries of what philosophers do qua philosophers are too fuzzy to permit us to establish such a claim. I have tried to suggest, via a few examples, that often when philosophers think they are essentially concerned with word-meaning, they are really concerned with the meaning or interpretation of sentence-types, and that they can formulate their semantic hypotheses in terms of sentence-sized units, thereby avoiding the special difficulties one encounters in analyzing sentence-meaning into the semantic contributions made by the various meaningful constituents of the sentence. This suggests, in turn, one reason for not assimilating the job of the analytical philosopher to that of the lexicographer. It would still remain true that the question of this paper is crucial to the methodology of linguistics, even if not to the methodology of analytical philosophy.

¹ Philosophy of Language (Englewood Cliffs, N.J.: Prentice-Hall, 1964), pp. 34 ff

² This qualifier is meant to rule out such contexts as those in which the speaker is acting in a play or giving an example of a request.

³ This list of conditions is intended to be illustrative rather than exhaustive.

⁴ Philosophy of Language (New York: Harper and Row, 1966), pp. 154 ff.

⁵ Ibid.

⁶ This not to say that I cannot be justified in semantic claims without having established them in this full-dress fashion. If one could not know, or be justified in believing, many things without being able to give ideally complete justifications of them, we would be in a pretty mess in many areas of thought.

⁷ This is a less ambitious task than that of determining what sense the word has in each of these contexts, although the lexicographer always, and the philosopher sometimes, asks a question like this in the course of trying to answer the more ambitious question.

⁸ I am using this phrase as an abbreviation for ‘ move rapidly by springing steps so that there is an instant in each step when neither foot touches the ground’.

⁹ This is not to deny that any judgments we make about the adequacy of a total system are highly fallible, nor that given any such system we are likely to construct, it is highly likely that a better one could be constructed.

¹⁰ Of course one might say that the conclusion, e.g., that ‘run’ has different senses in (13) and (14) depends on the assumption that ‘ run’ contributes at least part of the difference in the meanings of the two sentences. But this is just our conclusion over again.

¹¹ See his article ‘Meaning’, reprinted in this volume, pp. 53 ff.

¹² ‘I Know What I Want’, Proc. Arist. Soc., N.S. LVII, 1956-7.

¹³ Dreaming (London: Routledge and Kegan Paul, 1959), p. 80.

¹⁴ This entailment is itself highly disputable, but I shall accept it for the purposes of this illustration.