Methodological Relevance of Language Models with Expanding Sets of Sentences

Abstract

One of the very fruitful notions introduced in the theory of generative grammars is that of deep structure. The deep structure of a sentence, as distinguished from the surface structure, has to reflect the relation of a given surface structure to structures which underlie it in the derivation process. The essence of generation consists in a recursive enumeration of the set of all possible sentences of a language. Recently, some followers of Chomsky tend to understand generation in a different way — as a synthesis of a set of sentences departing from the given sense. Therefore, the notion of deep structure was identified with the semantic structure of a sentence. This identification was facilitated by some views expressed by Chomsky himself. Already in 1959, in his review of a book of Skinner (in Language 35(1959)), Chomsky criticized the behaviorist conception of the language acquisition in the childhood, and suggested as an alternative the hypothesis of the innate character of language categories. This rebirth of Kantianism has been criticized already by some scholars (for instance, by H. Hiz and Mrs. EL Charney) stating that it is not a necessary implication of the Chomskian theory.