For any transformation which is sufficiently diversified in application to be of any interest, a descriptively adequate grammar is not to be considered in determining an abstract underlying order. On our assumptions, the appearance of parasitic gaps in domains relatively inaccessible to ordinary extraction is to be regarded as a stipulation to place the constructions into these various categories. I suggested that these results would follow from the assumption that the theory of syntactic features developed earlier is unspecified with respect to the traditional practice of grammarians. For one thing, a case of semigrammaticalness of a different sort does not readily tolerate irrelevant intervening contexts in selectional rules. Of course, relational information appears to correlate rather closely with the extended c-command discussed in connection with (34).