On our assumptions, relational information delimits the strong generative capacity of the theory. To provide a constituent structure for T(Z,K), this analysis of a formative as a pair of sets of features does not affect the structure of a corpus of utterance tokens upon which conformity has been defined by the paired utterance test. Clearly, this selectionally introduced contextual feature cannot be arbitrary in the ultimate standard that determines the accuracy of any proposed grammar. We have already seen that the notion of level of grammaticalness is not subject to problems of phonemic and morphological analysis. From C1, it follows that most of the methodological work in modern linguistics is rather different from the traditional practice of grammarians.
see also: WikiPedia -- Chomskybot