On classifying semantic features

The semantic analysis of natural languages rests crucially on at least the following two assumptions: (i) the meaning of a given sentence can be accounted for on the basis of the words or, more precisely, the dictionary entries of which it consists, and the syntactic relations connecting these items; (ii) the meaning of dictionary entries are not unanalyzable wholes, but can be decomposed into elementary semantic components. These two assumptions are, of course, closely related to each other. The internal organization of the meaning of dictionary entries must be of a form which determines how they enter the composite meaning of more complex constituents according to the syntactic relations within these constituents. The syntactic relations in turn must be specified in such a way that the correct combination of the meanings of related constituents can be determined. A first attempt in this direction has been made by Katz and Fodor (1963) and Katz and Postal (1964). It is based on the assumptions that the syntactic relations in question can be defined in terms of underlying or deep phrase markers as specified in Chomsky (1965), and that the meaning of dictionary entries as well as of more complex constituents is given by strings of basic semantic components. The latter assumption was only a first approximation which turned out to be far too simple. It has been changed rather radically in the meanwhile.¹ In the present paper I shall concentrate on certain problems deriving from assumption (ii) above, more precisely on some aspects of the nature of basic semantic elements. It follows from the introductory remarks that even a discussion of such problems must keep in mind the important interdependence of assumption (i) and (ii).

The set of basic semantic elements has been divided by Katz and Fodor (1963) into two types of elements, called semantic markers and distinguishers. Although Katz has defended this distinction recently (Katz, 1967), it seems to me that it is an outcome of a rather early stage in the development of a semantic theory and must be rejected as theoretically unmotivated. I have presented my arguments for this rejection in Bierwisch and will not repeat them here. We are left then with only one type of basic semantic elements which might be called semantic features. These primitive terms, from which semantic descriptions of natural languages are constructed, do not, however, form an unstructured, amorphous set. Rather they are classified into several subtypes according to certain aspects, thus constituting ultimately a highly structured system of underlying elements.

¹ See Katz (1964a, 1964b, 1966, 1967). For a reinterpretation of several of Katz’ proposals see Bierwisch (1969). We shall discuss some of the changes involved in more detail below. Weinreich (1966) has sharply criticized the original assumption of mere concatenation - or logical conjunction, for that matter - of semantic components. His counterproposal of distinguishing clusters and configurations of components does not work, however, for reasons briefly discussed in Bierwisch (1968).