¹ This work was supported in part by the U. S. Army (Signal Corps), the U. S. Navy (Office of Naval Research), and the U. S. Air Force (Office of Scientific Research, Air Research and Development Command), and in part by the National Science Foundation.

I wish to thank Hilary Putnam for serving as a spirited ‘devil's advocate’.

² Cf. H. Putnam, Some Issues in the Theory of Grammar, Structure of language and its mathematical aspects (proceedings of Symposia in Applied Mathematics) 12.25–42 (1961).

³ This is suggested by the fact that the entries in good dictionaries are often inadequate for philosophical words. One can check this by consulting the O.E.D. under the words cause and number.

⁴ Notice, however, that, assuming the problem is genuine, Ziff's way out, though it allows him to get on with his theory construction, does not solve the problem. A solution requires a way to determine what the sample is a sample of, otherwise even after analysis we are not in a position to say whether we have established the meaning of m_i in English, in a dialect of English, in a certain idiolect of a particular dialect, or perhaps in some bizarre population cutting across these natural divisions.

⁵ That a method of semantic analysis should apply to any word or morpheme in the vocabulary of the language under investigation is clearly a condition of its adequacy. The degree to which there are words to which it fails to apply is the degree to which it is incomplete. That a method of semantic analysis should apply to any language is a highly desirable feature: a semantic theory which applies to any language is fully general and hence more interesting and more easily testable than one which does not apply to some languages. (More interesting because it reveals more about the nature of language, i.e. gives more insight into linguistic universals ; and more testable because the theory can be tested against a wider and more diverse set of linguistic structures.) Nevertheless, one may wonder why a method of semantic analysis should be constructed to apply also to dialects and idiolects. There are various reasons why the same method of analysis which applies to languages should equally well apply to dialects and even idiolects, not the least of which are these. We can expect that some words will differ in meaning in different dialects of the same language, and since we want to know which words these are and what the differences are, we will want a method which applies to dialects. And for reasons of economy and confirmation it is best if the method is the same one which is utilized in the study of languages. In the case of idiolects also we can expect differences in meaning for certain words, especially in philosophy and other specialized discourse where usage often departs radically from ‘ordinary language’. Again we will want some method to determine that words differ and what the differences are.

⁶ Cf. C. G. Hempel, A Logical Appraisal of Operationism, The validation of scientific theories 52–67 (Boston, 1954) ; Fundamentals of Concept Formation in Empirical Science, International encyclopedia of unified science 2.7.20–50 (Chicago, 1952).

⁷ This is as it should be. If SA were shown to fail as an evaluation procedure for dictionary entries simply because its input of utterances was inadequate, it would have been shown to fail for essentially trivial reasons. Furthermore, there is no reason to make the assertion D_sa relative to the input of pairings. But why this is so must be explained after we have explained the nature of such pairings.

⁸ Ziff spends some time on the distributional basis for such segmentation, but I see no reason to enter this discussion here.

⁹ Cf. H. Herzberger, The Decision Problem for Transformational Grammars, National Science Foundation, Transformations and Discourse Analysis Project, Paper 33 (Philadelphia, 1961).

¹⁰ The late English philosopher John Austin is responsible for the ideas behind this approach to semantic problems and for the concept of a speech act.

¹¹ According to Ziff (171), m_i has meaning in L if and only if Cm_i has nonnull members.

¹² This is my term, not Ziff's.

¹³ Boolean functions are those that can be obtained by starting with constants and the identity function and applying a finite number of times only the operations for obtaining the complement of a class, the union of two classes, and the intersection of two classes.

¹⁴ There is a printer's error in Ziff's book on page 155. There this set is characterized as āb + āb, but it is clear from the text that āb + ab̄ is the intended characterization.

¹⁵ Because these notions are not essential to our discussion, I will not try to describe them. Cf. 158 f.

¹⁶ Ziff says (193), ‘that the final step (forming Cm_i) consists in formulating a dictionary entry or something approximating to such an entry.‘ But Cm_i is a set of sets of conditions, not an entry. And since the point at issue is whether the system of analysis set forth constitutes even an evaluation procedure, we need not bother with the question of how a dictionary entry or something approximating one might be constructed on the basis of Cm_i.

¹⁷ This point was suggested to me by a remark of Noam Chomsky's.

¹⁸ It might be argued that when the expression language and the language under study are the same, the system of taxemic-semantic correlations is given automatically. However, this reply would merely beg the question. Each branch of linguistics aims at uncovering and explicitly stating the rules which explicate the skills and abilities of the native speaker that enable him to do something essential in the use of language, and hence the adequacy of a theory on any level is to be judged primarily by how well its rules accomplish this task of reconstruction. The more closely a linguistic theory matches the performance of the native speaker, the more satisfactory it is.

One of the main dangers encountered in formulating such rules is that they may be constructed in such a way as to be workable only when an appeal is made to linguistic intuition. By this is meant that in order for the rules to be used for the purpose they are intended, it is necessary that a fluent native speaker use his linguistic abilities to guide their application. This constitutes vicious circularity: the rules are supposed to reconstruct the native speaker's abilities and yet they cannot perform their function without the native speaker using these abilities to apply them. As much of the ability and skill of the speaker as are required to apply the rules, this much at least the rules fail to reconstruct. And to just this extent, the theory containing them is empty. Hence, this reply flagrantly begs the question because it construes Ziff's theory so that it appeals to just those skills and abilities that his theory is supposed to reconstruct.

¹⁹ The multiplicity of conditions associated with an utterance type in the empirical set will be further multiplied if the utterance type is ambiguous on the semantic but not on the grammatical level, i.e. if it exhibits the type of ambiguity which cannot be marked by the grammar but requires semantic categories to mark it, for example, The boat is light, The bill is small, Watch the door. To mark such ambiguities requires semantic categories, which are not available in the process by which regularities pertaining to utterances in the empirical set are established. Usually, in the environment where such utterances occur, various conditions determine which of two or more interpretations fits; but, at this stage, an appeal to specific conditions to decide which interpretations pertain to which tokens is viciously circular, since it is just these conditions that such decisions are intended to reveal.

²⁰ Appeal to the concept of a ‘speech act’ is wholly useless in this regard. No theory of speech acts exists and thus we can only tell what speech act has been accomplished after understanding the utterance token in terms of features of the context in which it occurs. That is to say, we determine the kind of speech act that has taken place on the basis of a knowledge of what conditions obtain in the utterance-context. To know that the occurrence of a token of the utterance type The door is open is employed to perform an act of ordering or commanding rather than one of asserting or commenting, it is necessary to know such things as that there is an open door that can be closed by the person addressed.