Word-sense

In spite of the inconclusiveness of some of the answers to points (α), (ß), (y) , and (δ) I now make a modest start with ambiguity and the theory of dictionaries. Its modesty must be emphasized. What exactly is involved in a satisfactory explanation or specification of the intended truth-condition for a sentence s? Satisfactoriness in an explanation is relative to an interest. What is the best formulation of the theoretical interest we should have? Till more thought has been trained on this question we have no very clear idea (cp. note b, p. 19). And what are we to make of the fact that most words don’t really have dictionary definitions (certainly not intersubstitutable equivalents)? As Putnam has recently stressed, most natural-kind words do not.

Webster and Larousse have large numbers of pictures, especially for such natural kinds as plants, animals, birds, etc. In the Politicus Plato approved such explanations, and its serves a clear purpose which nothing else would serve to start off by fixing the extension of a term in this way - ‘ whatever belongs to the same kind (e.g. animal species) as this. . .’ [here a picture or display is supplied].¹ But is this the sort of thing our dictionary is meant to do, or not? Having raised this problem I abandon it and try to draw out the consequences of such clarity as we do have about the minimal requirements on a dictionary.

A dictionary entry will presumably specify the part of a word and thereby determine its grammatical compatibilities in the terms of the best grammar for the language. (If a word can figure as different kinds of speech this will immediately give rise to more than one entry for it, and by our proposals for sense counting it will then rate as ambiguous from the outset.) Next, the dictionary will or may specify a defmiens or lexical citation for the meaning of the word. The first point to be made about this we owe to Aristotle (Metaphysics Z).

Consider the word ‘snub’ in ‘ Kallias is snub-nosed’. If we wrote the entry for ‘snub ’ as ‘ (adjective) concave ’ then we might seem to have made it possible for ‘ this mirror is snub’ to be true. But if we write ‘(adjective) concave-nosed’ then it would look as if when we go back to our original sentence and write its truth-condition it must be ‘ Kallias is concave nosed-nosed’.² Nor can the trouble be cured by defining ‘ snub-nosed ’ in one piece. It would not cope with ‘She had a charming nose, almost snub’ or ‘A retrousse nose is something like a snub nose’. Consideration of Aristotle’s example suggests that at least some dictionary entries will have to carry another indication like this: ‘Snub: (adjective) [of a nose], concave’. We may call this extra component [. . . . . . . .], which imports what is really syntactic as well as semantic information, a directive, and call the other component the analysis or citation for the word. It is blatantly obvious that we are not yet in a position to explain directive otherwise than as ‘what cannot find a place in the analysis itself’. And it is most unlikely that the need for the [. . . . . . . .] device always arises in the same way. I expect that the point about snub may seem rather insignificant. It is not, however, insignificant and it bears on a number of disputes which would have benefited by exposure to it.³

Consider and and but,⁴ or aliquis (someone) and quidam (a certain person, the speaker not letting on who), or even perhaps gaggle and flock. If we try to account for the differences in each pair by making different analyses or citations do the explanatory work, then we reach some absurd positions. But and and differ in meaning. Suppose we account for this by a difference in citation. Then the sentence His speech was long hut impressive will count as false even if the speech was both long and impressive if there exists no such contrast as the one which the different citations for and and but require. Finding this absurd, we may feel obliged to accept one and the same simple truth-functional citation for both words. And we may wrongly reject the idea that but and and have any difference at all in strict meaning. Note, however, that we shall be doing this purely and simply because their lexical citations are identical. Frege boldly took this course but it is equally counter-intuitive. Aristotle’s point shows that this reason is not a good reason to identify their meanings and that there is another option. The generalization of the distinction between a directive and a citation makes it possible for us to see how and and but have different lexicon entries and differ in meaning without differing in citation. The standard (sometimes called rhetorical) difference between them comes in their respective directives. But this leaves the difference within the realm of semantics, and we must adjust the statement of the Fregean semantic programme to find room for such effects of directives.

The point may still seem a trivial one. Perhaps I can make it seem more interesting by suggesting that it also solves one small problem about conditionals.⁵ Philosophers have been puzzled about the part played by if with the subjunctive mood. Subjunctive conditionals seem to be strongly connected with contrary-to-factness. Moreover they seem to be connected by something semantic - or if not something semantic then something very general. On the other hand it is noticed that the truth of the antecedent p in if p were true then q would be true could hardly be held to falsify the conditional. A part of the answer to this conundrum is surely that a dictionary might well abandon any attempt to build counterfactuality into the citation for subjunctive if and still account for what needs to be explained by writing an entry on these lines: if [used with subjunctive when the speaker is either speaking as one believing the untruth of the antecedent or as one not believing the truth or untruth of the antecedent] ... If parties to disputes about subjunctive if had noted the rather uninteresting case of snub and seen that not all meaning specification can be by provision of an intersubstituend many hours might certainly have been saved.

Even with so small a fund of information as we have about the theoretical requirements on a good dictionary, there are one or two further conclusions which follow with something near certainty.

Sometimes it will happen that, because one sort of occurrence of a word has suggested one definition and another sort of occurrence has suggested another definition, we have a prima facie case for ambiguity of the word. But even if the two lexicon entries for the word seem to be sound and they yield correct readings for the respective types of occurrence with which they were intended to deal, this cannot prove that one general account could not have been given to cover both simultaneously. One might have been able to do better. So ambiguity is often difficult to prove. This is as it should be.

This will raise the question - ‘What will you do if someone takes an obviously ambiguous word and simply disjoins the two entries for it? Can the lexicon approach show there is anything wrong with such a concoction? And what is its test for detecting specious unity?’ Now certainly bank means either river verge or money depository. But this does not imply that it means either a river verge or a money depository. It is simply a brute fact which a theory of meaning has to explain, and not explain away, that the sentence I went to the bank this morning and wept many tears might on many occasions be entirely and completely ambiguous between two utterly distinct conditions. It is of its nature to prompt the question which do you mean, a river bank or money bank? Till we know which is meant we do not know what was said. Nor is it difficult to see that the disjunctive account would result in the conversion of actually true statements into false, e.g. All banks in the U.S.A. are now guaranteed by the Central Reserve Bank.

But of course there are unitary accounts of problematic words which are not put forward in a captious spirit and which are much more difficult to adjudicate. And notoriously there is the word ‘ good’, for which such unitary definitions have been advanced as satisfies the criteria (or comes up to the standards) prevailing for the evaluation of items of its kind or answers to the relevant interest. Of course these definitions have been offered in the pursuit of philosophical enterprises with which I am in general sympathy. For this very reason I am anxious to show that they go too far and that the Fregean lexicon approach is stringent enough to be committed to their actual inadequacy. The example of ‘good’ has the advantage of raising a large number of general problems about word-sense, so there is some point in a rather extended discussion. Since the arguments against both proposals are broadly similar, I shall choose only one of them, the second having the attraction of being interestingly defended by Paul Ziff.⁶

In the first instance Ziff’s proposal might be put in something like this form - good: (adj.), [used in context of evaluation with some determinate evaluative interest. Let the interest in the case be interest i,] answers to the interest i. Could anything on these lines possibly be adequate? There is certainly something which the proposal explains. What it explains, however, is what I originally labelled a type (iii) question, not a type (i) question. I do not think it fully explains the utterance conditions of all, or indeed any, sentences containing the word good.

Consider the sentence She has good Iegs.^a It has a number of interesting properties, of which the most striking is that it is a straightforward counter-example to the notion which has first occurred to many philosophers who have reflected on goodness --that to any determinate thing of a determinate kind there is annexed (anyway for a determinate speaker or hearer) a determinate criterion of goodness for the goodness of that kind of thing. For the trouble with legs is very roughly this - that they can be assessed aesthetically as good = beautiful, they can be assessed functionally or ‘ instrumentally ’ as good — strong or well suited to this or that activity, and they can be assessed medically as good = not maimed, healthy. The undoubted connexion (to which I shall come in a moment - this is the type (iii) question) between these dimensions of assessment does nothing to establish their identity. There are other dimensions of assessment for legs too, but it is interesting (or anyway it is to be noted) that the verdict of goodness in these other dimensions has then to contain express reference to that special dimension. She has legs to show to Dr X (who would be interested in such and such an anatomical feature). She has good legs for doing such and such a test on. Her legs would he good to eat.

How, if at all, does this threaten the unified account of good? The unified account could be correct if good combined with legs to produce one set of conditions; or perhaps it might be correct if there were an indefinite or potentially infinite number of such conditions corresponding to a supposably indefinite number of logically possible dimensions of assessment for legs. (Goodness would then be a very indeterminate property.) Neither situation obtains. The number of interpretations sticks obstinately at three or four and I submit that we know what they are in advance of any context of utterance. It may be said that it is in the nature of legs to fix and limit to three or four these three or four interpretations for good legs. This is true but not to the point, if there really is a genuine indeterminacy between three and only three genuinely distinct genuine interpretations of the sentence She has good legs. It is pertinent to add that when Von Wright wrote a book called The Varieties of Goodness he found that the pattern good+ noun (with ‘good’ unqualified) was associated with some six or seven categories of assessment. Not many more. Not many less. This did not result from his restricting his choice of values for the noun-place. What he discovered does no doubt result from the fact that as a fact of natural history some six or seven dimensions of assessment, instrumental goodness, medical goodness, the beneficial, the useful, hedonic goodness, etc., dominate all our standard evaluations. And no doubt there is something one can say to set them in a pattern. But if I am right this fact of natural history has infiltrated the semantics of the good. When good and some noun with which it is joined coalesce to produce their joint output, good carries something more distinctive to the compound than the colorless idea of evaluation per se. What it carries depends on the noun, but for some nouns there is more than one such distinctive meaning imported by the word good. It is this which embarrasses the unitary account, which might otherwise have been able to plead that it was fair enough to produce a unitary account by classifying nouns into some seven categories and giving a compound instruction for the production of truth-conditions according to the category of the noun combined with good.

It would be empty, I think, for the unity theorist to appeal to the possibility that contextual factors would always decide. In fact they do not. The remark She has good legs really is found to be indeterminate (and I should say ambiguous) in some contexts. And there are some where it would surely have to be found ambiguous. It is certainly vacuous to count the larger conversational context including elucidations of the remark itself as part of the context. By this proposal all words indiscriminately would admit of one account.

If I am right my contention can be reinforced by testing the answers to the relevant interests account the other way round. Suppose that I have been engaged to find a woman with legs about a foot shorter than the statistical average for women of her height. Suppose that Mr Y, an imaginative and powerfully backed film director, badly needs an extra with this physical peculiarity. If I find a woman with this characteristic then her legs answer the relevant interest. (Does she not herself answer it ? - which doesn’t make her a good woman.) She has good legs to show to Mr Y or good legs for my purpose or good legs/or the part of so and so in such and such a film. But none of these qualifications is detachable, and this is a linguistic fact which cries out for explanation. This awkward fact would also obstruct the attempt to give a unitary account of the word good which started from the schema good (noun) to/for (. . . . . . . .)with two slots and then explained good (noun) by the operation of some kind of ellipsis. The explanation of the admissibility of some ellipses and the inadmissibility of others would most probably involve a virtual acknowledgement of the ambiguity - theorist’s account of the word good. (In this connexion it is also worth remarking that the deleted ‘ to. . . ’ or ‘ for. . . ’ is extremely difficult to specify properly, with the required degree of vagueness, etc. Are good (= beautiful) legs precisely legs which are good legs to look at? And what is a good man a good man to or for?)

These difficulties in the unitary account of ‘good ’ lead to another difficulty in Ziff’s view. In the case where we have a substance of kind X called a good X and there is one and only one interest or set of standards which is agreed to be relevant for being a good X simpliciter, there is little or no trouble in meeting the ancient difficulty which troubled Stevenson and Ayer about evaluative disagreement. If disputants understand what an X is then in this sort of case that does simply fix a common interest to decide the question whether X is good or not. But where there is no clear and agreed interest, or where the whole substance of dispute is ‘which interest is the relevant interest? ’ matters are not so simple. Examples tend to bunch in areas of some importance in moral philosophy, e.g. ‘ good man ’, ‘ good plan ’, or ‘ good thing to do ’. A disagreement about what to do need not always, even between partners with a common interest such as a man and his wife, be a disagreement about means to ends. A disagreement may be a disagreement about which interest or end is the relevant interest or end (all relevant questions of means being relatively easy in the case, let us suppose). This is quite compatible with its being a disagreement about whether X is or is not a good plan, or a good thing to do. What seems to be wanted is to transfer the word ‘ relevant ’ from the directive to the citation or analysis, thus transforming the analysis itself into answers to the relevant interest. But this proposal can only aggravate all the difficulties of unifying the sense of good. The dictionary has still to say more than it can by means of one entry about what interests are the ones that qualify, and about which interest is the relevant one in a given case. Suppose scaring birds is the relevant interest. Then in dressing a scarecrow a used and worthless hat will come to count as a good hat. A good hat simpliciter, not just a good hat to give to a scarecrow. But don’t we want to hold onto the idea that a bad hat is a good hat to give to a scarecrow? (And see the remarks above about the elliptical analysis. It is possible, I suppose, that we could altogether drop the nonelliptical good X (simpliciter) locution and still say most of the things we want to say. But all I am trying to do is describe how good works at present.) Nor is it clear that there will always be anything one would call the relevant interest. Having got himself a ploughing ox and a house, Hesiod wants a wife. Is good cook the relevant interest, or good housekeeper, or good gardener} (Or good legs?) Or good woman even? (Which is not, I am supposing, some unified logical product of the qualities just enumerated.) Even for Hesiod there is no such thing as the relevant interest for good wife. Nor is it any use to go over to answers the relevant interests (plural) and say the context or the item in question determines which they are. It does not always. Hesiod needs to be able to hold onto the option of saying they are good legs (beautiful) but not strong, which is what above all else any wife (or ploughing ox) of his will need and must have.⁸

We are not yet finished with good and shall come to back it when I finally get to vindicate the claims I made at the beginning about three types of question. But first another methodological doubt.

So far so good, it may be said, but the lexical test of univocality, counting the several entries you find you have to give for a word, gives only a sufficient condition of ambiguity. It cannot give a necessary condition of ambiguity (or sufficient condition of univocality) because a lexicon entry may itself be ambiguous. (Cp. Aristotle, Physics 248 b.) We have said nothing at all so far about what makes an account of a word-sense one account. In fact, it may be objected, this whole view of ambiguity can only be correct by being circular. I will arrange my pleas against this objection in order of ascending strength.

The first reply is that even if the account is circular it is at least true and, what is more, can serve the sort of purpose many circular necessary conditions very usefully serve. (Think of the transitivity of identity.) Suppose we have, say, a particular reflexive occurrence O₁ and a particular interpersonal occurrence O₂ of the word ‘deceive’. Then the above explication of ambiguity tells us that the existence of non-interchangeable paraphrases for O₁ and 0₂ is not a sufficient condition of the ambiguity of ‘ deceive ’; that a necessary condition of univocality is the discoverability of a non-disjunctive account to cover both O₁ and 0₂ (we have already shown why this is a non-trivial requirement); and that it is a necessary condition of the ambiguity of ‘ deceive ’ that one should be unable to contrive a single adequate account of the contribution of ‘ deceive ’ at both O₁ and 0₂. In practice this is quite enough for many discussions.

The second reply is that there is a little more we can supply about what makes a lexical entry into what is properly one entry. Apart from the test by which I failed a speciously unitary account of ‘bank’, there are other tests. Suppose that it is possible to state a principle to assign all occurrences of a word w to one or other (or a third or. . .) of two (or more) disjoint classes, and that a unitary account is then produced which gives it meaning S₁ in environment V₁ and S₂ in environment V₂. . . Consider for example the definition:

Sharp (adjective) if a noise then shrill and if a curve then abrupt and if a knife then having a fine edge. . . and suppose (what is not actually quite true) that the requirement can be satisfied that the different sorts of environment exactly partition the different senses.

The definition is very perverse and a correct one ought presumably to make use of some directives. But how can we fault it? It seems clear that there is a principle of learn-ability and indeed one matter of fact on which the definition fails. To teach the sense of ∅ involves putting anyone who knows what not means into a position to understand not ∅. And surely someone who knows what sharp knife means might understand the sense of sharp, and might also know what X is not sharp means, without understanding the latter as either both a noise and not shrill or both a curve and not abrupt or both a knife and not having a fine edge. And surely saying a knife is not sharp is not saying this ?⁹

There is a third answer to the objection, and to the suggestion that however far one gets in semantics one may still be serving up ambiguous truth-conditions. Like every other science semantics can only be judged by whether it explains what needs to be explained, and how well. What needs to be explained in the present case is our capacity to understand sentences. The notion of ambiguity only comes into the business at all because it is needed to explain this understanding. Since it has no other rationale in any case, there are no sensible doubts about it which cannot be converted into clear doubts (which are more sensible doubts than my objector’s) about whether English speakers’ understanding of English sentences has been explained.

I shall conclude by trying as promised to distinguish three questions which people have asked about ambiguity. Everyone who has felt the pull of the unitary analysis of ‘ good ’ will wish to ask whether my argument was meant to show that good is a ‘ mere homonym’ with something like, say, seven senses. Or twelve senses? If one simply said that and simply called good a ‘homonym’ tout court that would seem to be a mad conclusion. Aristotle, who believed good was not univocal, suggests in the Nicomachean Ethics (I. 6) that its senses are either ‘analogically’ or ‘focally’ related.¹⁰ He does not work out the suggestion but I shall tentatively try to do so now. Ziff’s analysis of good has the virtue of marking something which the different lexicon entries for good would certainly have to have in common, viz. the notion of something’s satisfaction of some definite interest or other; but, pace Ziff, the lexical analysis of good cannot dispense with determinate specification of what sorts of interest. So it might be suggested that all the lexicon entries are structurally analogous to one another. That echoes one of Aristotle’s suggestions. Even more tentatively I suggest that we may try to account focally for the particular array of interests represented in the separate entries for good X, but by an anthropological explanation which finds a pattern in the interests which have found their way into the several senses of good kept strictly separate from one another by the lexicon. What partly organizes the pattern of interests picked out by the users of the language is what the users of the language conceive that they want and need in the general framework of the life they desire for themselves (eudaimonia). If it fits anywhere it is here that the piece of information that ‘good’ is the most general word of commendation in English comes in. (But neither this piece of information nor Ziff’s account is a recipe for constructing a single lexical analysis of good.)

This is very tentative and nothing hangs on the details. The generalizable point is that when we distinguish senses we do not necessarily condemn a word to pun status, and that it is a part of the equipment of a mature speaker of a language that he possesses some principles (e.g. extension by analogy, extension of senses from a focus) which enable him to invent or ‘cotton onto’ new uses of a word and see their rationale. The new uses which do get into the dictionary should be kept apart there from one another and apart from the original use. For this very reason it is not necessary, and at any moment of time it will never be desirable or possible to put all of the new uses in the dictionary. (Consider, every week somebody may invent yet another use, technical, erotic, penal, topological or far-fetched, for, say, the word ‘inside’.) But if and when uses do get sufficiently standardized to merit a place in the dictionary, they will give rise to new and distinct entries. New uses get into the dictionary when they settle down into distinct and distinctive senses which are worth noticing in the dictionary, and they only arrive there because they make some recurrent difference to truth-conditions. It is difference to truth-conditions of sentences, not type (iii) connexions, which must primarily concern the dictionary.

The account I have offered of ambiguity is in some ways rather remote from traditional discussions, which concentrated on paraphrases for particular words. Paraphrase is the method employed by Aristotle in his, in my opinion, defective proofs of the non-univocality of good. (See Nicomachean Ethics I. 6, the earlier Eudemian Ethics i. 8 and his fullest early discussion of the matter, Topics 107.) Traditional discussions find it enough to do what I think is not enough for a type (i) question. They point to the fact that a word like good can be paraphrased in completely different ways in its occurrences in such contexts as good food and good knife and good lyre. At least in the last twro, however, it seems obvious that the word good itself makes exactly the same input to the sense of the sentence. But it is also obvious that the question of sense is not necessarily the question to which the traditional discussions were directed. It takes no great charity to suggest that they are really interested not in the input but in the output of a word, the part of the paraphrase of a whole sentence for whose presence some particular word in the sentence is responsible. In fact they are interested in what one might call proposition-factors. There is no difficulty in constructing a diagram analogous to the p. 24 diagram where (1) assumes as a new value a word type with a sense, e.g. good, (2) has as its new value a word-token of that word type, e.g. good as it occurs in this is a good knife and (4) takes the value has a fine edge suitable for cutting, this being one component of the paraphrase This knife has a fine edge suitable for cutting, which paraphrases This is a good knife.

It may be asked, why should anyone interest himself in these output questions. I think the answer is that type (ii) questions sometimes constitute interesting problems about property-identity. Surely, it has been thought, goodness in a knife is not the same property as goodness in food? And not even the same as goodness in a lyre? And the inferential consequences of the possession of one sort of goodness will be very different from those of another. (I can imagine that similar questions about the relation of input and output might matter for the discussion of some implications, e.g. ‘ought implies can’.)¹¹

The sign that this may be the best view of the apparent conflict between the approaches, and that more clarification is required than resolution, is provided by the fact that Aristotle’s alarm in Physics, Book vII about definitions themselves turning out to be ambiguous looks most comical when he suggests that double or much may be ambiguous. (The worry is that much air is not comparable with much water.) The reason why this looks like a senseless anxiety is that there is neither an input question nor an interesting output question here. Much is not a first order property at all and a fortiori it is not a first order property about which there can be questions like questions about goodness in lyres and goodness in knives or food. There is no interesting or anyway straightforward type (ii) question to be asked about it.

¹ In general it will be very difficult to determine in advance what will be relevant and what irrelevant to the collection of outlying members of the class or to testing the claims of founder members. Consider the criterion now employed in zoology to determine the bounds of the Carnivores. Fusion of the scaphoid and lunareis hardly part of the sense, perhaps, or anyway the intension of ‘carnivore’. Kind as it figures in this sort of definition schema is theory-laden. Or rather it waits on whatever may be the best theory of the relevant kind of kind.

² It is sometimes objected to Aristotle that the objection to ‘ concave nosed-nosed ’ is (at very best) grammatical. Consider then fetlock = horse's ankle and then use this paraphrase to rewrite ‘the stone hit the horse’s fetlock’ as the ‘ the stone hit the horse’s horse’s ankle ’. If this were grammatical it would itself be ambiguous! It does not rule out that mad reading of the genitive which suggests that one horse owned another horse which owned an ankle. Para¬ phrase is meant to eliminate, not to introduce, ambiguity.

³ Indeed it is part of what is a much larger flaw in a programme of philosophical analysis, the unremitting search for analysis by intersubstitutable equivalents, which has taxed the ingenuity of a whole tradition, Plato (Theaetetus), Aristotle, Leibniz [Characteristica Universalis), Frege, Moore, Wittgenstein (Tractatus), early and middle Carnap and even some Oxford philosophy. When Goodman, Quine and Carnap lower their demands to some looser equivalence I think that this smacks more of disappointment (and of Goodman’s and Quine’s dissatisfaction with the notion of a priori necessity) than of any fundamental reappraisal of what is involved in giving an adequate explanation of sense. Once we revise our ideas about what this really involves it is possible again to see Goodman’s merely extensional requirement as much too permissive. And a number of other problems are transformed too. We can take it in our stride, for instance, that different definite descriptions can fix the sense of a proper name and all in their different ways fix one and the same sense.

I hold no brief for Austin or his champions such as Stanley Cavell on the meaning or appropriateness of ‘Did you tie your shoes voluntarily?’ (see Austin’s ‘Excuses’ in Philosophical Papers (Oxford, 1961)), but parties to the dispute ought, I think, to reflect whether any inappropriateness there may be in the question derives from an implicature of reluctance, or from a directive for ‘voluntary’ such as: [used when the question of willingness arises].

⁴ But and and as used internally within whole sentences. Their use as particlesto link different sentences, paragraphs, etc. is not in consideration here. The preface paradox (‘ I believe every proposition in this book but I am confident that a mistake will be found somewhere’ - which rules out the possibility of seeing the book even notionally as a single conjunctive assertion) makes me doubt that the two uses are interreducible.

⁵ Nor can the difference between citation and directive be irrelevant to the many disputes there have been between cognitivists, emotivists, prescriptivists and others in ethics. If the differences between ‘Limey’ and ‘Englishman’ or ‘nigger’ and ‘negro’ are differences in something like directive then even the best putative examples of ‘ emotive meaning’ show much less than anti-cognitivists have wanted them to show against the naturalists’ theories of mean¬ ing. But I must remind the reader at this point that directive was defined negatively. Work would need to be done, and the idea would need to be broken down, before more use could be made of it in ethics.

⁶ Semantic Analysis.

⁷ An example I have several times discussed with Michael Woods, to whom I am greatly indebted.

⁸ I think that it is sometimes suggested that the information we apparently need to get into the dictionary in order to overcome these difficulties could be stored under the substantives which constitute the different values of X in good X, instead of distributing it between several different entries for good. I am not sure how this could be made to work but, whatever else it does, presumably it ought not to account for the ambiguity of she has good legs by necessitating the hypothesis of an ambiguity in legs!

⁹ Another example, also Aristotle’s, is white. For every hundred people who understand the usual sense of the word white there can only be one or two who know what white noise is. Does this really impair their general understanding of not-white?

¹⁰ For the distinction see G. E. L. Owen in Dühring and Owen (editors), Plato and Aristotle in Mid Fourth Century (Gothenburg, i960), also Austin, Philosophical Papers, pp. 34-41 (Oxford, 1961). One of Aristotle’s favorite clear cases of focal meaning, healthy, happens to provide a particularly good example of failure of unitary analysis. It is exactly similar to the failure with good. The only way of representing healthy as unitary is to write the lexicon entry as e.g. pertaining to the health or well-being of an organism. This collects up healthy complexion, healthy constitution, healthy weather, healthy climate, healthy drink all right but only at the cost of allowing such oddities as healthy hospital, healthy cure, healthy lecture, healthy text¬ book. Pertaining to is too vague. But to specify something more exact involves splitting the lexicon entry for the word. The true account of the matter is that the various senses of healthy are related by being arranged in different ways around the ‘focus’ health. It is no accident that all the things which are properly called healthy are so called but the ways are different ways. And this explanation has nothing to do with analogy.

¹¹ This difference between input and output questions may be one of the things at issue in one so far fruitless dispute between Fred Sommers and his critics (conveniently summarized in Jonathan Bennett, Journal of Symbolic Logic, September 1967, p. 407).