The theoretical status of the syntactic markers in KF is not clear. It is probably fair to understand that the function of the syntactic marker SxM_i is to assure that all entries having that marker, and only those, can be introduced¹ into the points of a syntactic frame identified by the category symbol SxM_i. In that case the set of syntactic markers of a dictionary would be just the set of terminal category symbols, or lexical categories (in the sense of Chomsky 1965), of a given grammar.

It is implied in KF that this set of categories is given to the lexicographer by the grammarian. Actually, no complete grammar meeting these requirements has ever been written; on the contrary, since KF was published, a surfeit of arbitrary decisions in grammatical analysis has led syntactic theorists (including Katz) to explore an integrated theory of descriptions in which the semantic component is searched for motivations in setting up syntactic subcategories (Katz and Postal 1964; Chomsky 1965). But before we can deal with the substantive questions of justifying particular syntactic features, we ought to consider some issues of presentation - issues relating to the form of these features.

In general, the size and number of lexical categories (traditionally, parts-of-speech) depends on the depth or ‘ delicacy ’ of the syntactic subcategorization. (The term ‘delicacy’ is due to Halliday 1961.) Suppose a category A is subcategorized into B and C. This may be shown superficially by a formula such as (2i); a Latin example would be ‘Declinable’ subcategorized into ‘Noun’ and ‘Adjective’. However, the specific fact of subcategorization is not itself exhibited here. It would be explicitly shown by either (2 ii) or (2 iii). In (2ii), A₁ = B and A₂ = C ; in (2 iii) [+ F] and [ - F] represent values of a variable feature² which differentiates the species B and C of the genus A. (An example would be ‘ Nomen ’ subcategorized into ‘ Nomen substantivum’ and ‘Nomen adjectivum’.) The feature notation has been developed in phonology and has recently been applied to syntax by Chomsky (1965).³

Single, global syntactic markers would correspond to implicit notations, such as (2i); sequences of elementary markers, to a feature notation such as {2 iii). The KF approach is eclectic on this point. The sequence of markers ‘ Verb → Verb transitive ’ for their sample entry play corresponds to principle (2iii); the marker ‘Noun concrete ’ seems to follow the least revealing principle, (2 i).⁴ To be sure, the examples in KF are intended to be only approximate, but they are surprisingly anecdotal in relation to the state of our knowledge of English syntax; what is more, they are mutually inconsistent.

A revealing notation for syntax clearly has little use for global categories, and we may expect that for the syntactic markers of dictionary entries in normal form, too, only a feature notation would be useful. In our further discussion, we will assume that on reconsideration KF would have replaced all syntactic markers by sequences expressing subcategorization.

Suppose, then, we conceive of a syntactic marker as of a sequence of symbols (the first being a category symbol and the others, feature symbols). Suppose the dictionary contains entries consisting of partly similar strings, e.g. (3):

This partial similarity could be shown explicitly as a branching sequence, as in (4):

But this notation, proposed by KF, does not discriminate between fortuitous homonymy and lexicologically interesting polysemy, for it would also produce entries like

KF would therefore have to be extended at least by a requirement that conflated entries with branching of syntactic markers be permitted only if there is a reconvergence of paths at some semantic marker; only, that is, if the dictionary entry shows explicitly that the meanings of the entries are related as in (6).⁵ But such

makeshift remedies, feasible though they are, would still fail to represent class shifting of the type to explore - an explore, a package - to package as a (partly) productive process: the KF dictionary would have to store all forms which the speakers of the language can form at will. We return to a theory capable of representing this ability adequately.

¹ By a rewrite rule, according to early generative grammar (e.g. Lees i960), or by a substitution transformation according to Chomsky (1965).

² A feature symbol differs from a category symbol in that it does not by itself dominate any segment of a surface string under any derivation; in other words, it is never represented by a distinct phonic segment.

³ I am informed by Chomsky that the idea of using features was first proposed by G. H. Matthews about 1957, and was independently worked out to some extent by Robert P. Stockwell and his students.

⁴ We interpret ‘Noun concrete’ as a global marker since it is not shown to be analyzed out of a marker ‘Noun’.

⁵ Even so, there are serious problems in specifying that the point of reconvergence be sufficiently Bow’, i.e. that file ‘record container* and file ‘abrading instrument’, for example, are mere hononyms even though both perhaps share a semantic marker (Physical Object). This important problem is not faced in KF, even though the examples there are all of the significant kind (polysemy). See Weinreich (1963a: 143) for additional comments.