Nonfiltering and Local-Filtering Transformational Grammars

Abstract

A transformational grammar delimits a set of grammatical sentences by virtue of containing (a) a base component which generates a set of base Phrase-markers (labeled trees or labeled bracketings), (b) a transformational component which maps these into derived Phrase-markers and (c) filter predicates which pose a further admissibility condition on the latter Phrase-markers. A string x is grammatical according to such a grammar if there is a Phrase-marker generated by the base component which is mapped by the transformational component to a Phrase-marker satisfying all the filter predicates and having x as its terminal string. Thus all these three parts of a grammar play a role in specifying what strings are grammatical — the base component by determining what Phrase-markers will be available for transformations to operate on, the transformational component by converting these inputs into derived Phrase-markers, the filter predicates by excluding certain of these outputs as inadmissible.

This work was supported in part by the 1969 Advanced Research Seminar in Mathematical Linguistics, sponsored by the Center for Advanced Study in the Behavioral Sciences, Stanford, California. The first author also received support from National Science Foundation Grant GS 2468.