this article I will argue, however, that this contrast between arguments and predicates is part of a more general pattern of contrast between referential and nonreferential phrases, in the sense of "referential" that has been shown by Comorovski (1989), Kroch (1989), Rizzi (1990), Cinque (1990), Longobardi (1991) to be relevant to extraction possibilities. I will argue that the contrast comes about because the relevant part of a nonreferential phrase is forced to reconstruct---to lower---at LF in order for the sentences to receive the correct interpretation. It follows from this analysis that LF lowering feeds Condition C of the binding theory, as suggested independently by Clark (1992). It will also be seen to follow that the behavior of predicates with respect to reconstruction does not constitute evidence for the necessary