“A Hungarian Answer to Locality: Zipper Merge”

Syntactic opacity can be selective, that is, movement landing in a matrix clause-position may be only grammatical from within a subset of embedded clauses. Existing work on selective opacity has primarily focused on the syntactic size of the embedded clause ( e.g. Williams, 2003). For instance, clauses with more left-periphery projections seem to be opaque to more cross-clausal movement types and lower positions in the hierarchy are harder to land in (Abels, 2012). This is Problem 1. In this talk, I use Hungarian to show that the left-periphery position of the embedded clause in the matrix clause also matters, that is, we have Problem 2: The higher a clause gets interpreted in the matrix left-periphery, the fewer movements are grammatical from within it. Since this is problematic for existing analyses, I provide a unified account for these two opacity-related problems.

Phase Impenetrability Condition (Chomsky, 2000, 2001) can be reformulated in terms of functional sequences: The heads crossed by movements must be increasing with respect to their place in fseqs. That is, any movement step crossing the highest head of a complement clause and then the matrix V head, will violate PIC. One strategy to avoid this is successive cyclicty, that is, stopping at a phase-edge and restarting the fseq-counting before crossing V. I propose that Hungarian uses an alternative method, namely, restructuring: The embedded clause gets extraposed and merges with the matrix clause whenever the matrix clause reaches its size, e.g. a matrix XP merges with an embedded XP producing a shared XP. (All XP projections must be identical with respect to features.) This means that the embedded clause’s fseq finds a continuation in the matrix clause’s fseq, therefore, no movement landing in this fseq-continuation will violate PIC. At the same time, before reconstruction/extraposition, positions along the matrix fseq remain unavailable (Problem 1 solved). Since the XP above the restructuring will be the maximal projection of the embedded X, and we can only move maximal projections, the embedded XP cannot be extracted and land in higher positions after it gets united with the matrix XP (Problem 2 solved).

Why may this be interesting on a more general level? On the one hand, partially shared functional sequences may have semantic consequences: Smaller embedded clauses will be semantically more dependent on their matrix clause. On the other hand, if restructuring is a last resort in the absence of phase edges, we expect that embedded clauses which are so small that no C head with edge-features projects can be “unlocked” only via restructuring (cf. Rizzi, 1982).