Generative Semantics

In recent works by such linguists as George Lakoff, James McCawley, Jeffrey Gruber, and Leonhard Lipka, a number of problems in the interpretive semantics model have been noted. These writers reject the idea that syntax should be the grammatical base. First, they say that syntax occupies an intermediate position between phonology and semantics. When using syntax as the base, it is therefore necessary to go in two different directions, up to the phonology and down to the semantics. Secondly, semantic features are universal whereas syntax and phonology are mainly language specific. They claim that it makes more sense to go from this universal base toward the more specific and divergent syntactic and phonological components of various particular languages. Third, they feel that there may not be any justification for a separate syntactic level at all, and if there is no independently justified level of syntax, then this level cannot be the base level since it doesn’t really exist. And fourth, as Wallace Chafe has pointed out, the speech act — and probably the origin of language as well — begins with concepts and these are encoded into language rather than the reverse being true. A final criticism of the interpretive semantic model is that it has two separate kinds of rules — transformations for handling syntactic and phonological matters, and projection rules for handling semantic matters.

The generative semanticists feel that only one type of rule is necessary and this is the transformation. They feel that they can handle semantic information by postulating higher predicates and predicate-lifting transformations which have the effect of combining a lower predicate with a higher predicate which results in a new word with all of the meanings of the two words from which it was formed. For example, they see the verb kill as consisting of the state dead plus the two higher level predicates came and become. Thus, taking the expression, cause to become dead one predicate-lifting transformation will change become dead into die, and a second will change cause to die into kill. It has been argued by non-generative semanticists that cause to die does not mean exactly the same as kill. The generative semanticists have answered that the higher predicate cause is different from the lexical item cause, and in fact differs in exactly those ways which make cause to die an accurate paraphrase of kill. In a way it appears that on this particular point the generative semanticists have come full cycle in their reasoning. They claim to be able to handle semantics without resorting to semantic features. Yet what they call their higher predicates look very much like semantic features since they make a distinction between them and regular lexical units or words.

Some of the best readings on the generative semantics viewpoint are James D. McCawley’s Where Do Noun Phrases Come From? in Readings in English Transformational Grammar (1970), Jeffrey Gruber’s Studies in Lexical Relations (1970), Paul Postal’s On the Surface Verb Remind in Linguistic Inquiry (1970), George Lakoff’s On Generative Semantics in Semantics: An Interderdisciplinary Reader in Philosophy, Linguistics, and Psychology (1971), and Leonard Lipka’s Semantic Structure and Word Formation (1972).