It may be, then, that this selectionally introduced contextual feature can be defined in such a way as to impose problems of phonemic and morphological analysis. Notice, incidentally, that an important property of these three types of EC is unspecified with respect to the strong generative capacity of the theory. Of course, relational information raises serious doubts about an abstract underlying order. I suggested that these results would follow from the assumption that the appearance of parasitic gaps in domains relatively inaccessible to ordinary extraction delimits the extended c-command discussed in connection with (34). Conversely, the speaker-hearer's linguistic intuition appears to correlate rather closely with the levels of acceptability from fairly high (eg (99a)) to virtual gibberish (eg (98d)).
see also: WikiPedia -- Chomskybot