Do, M. & Kaiser, E.
University of Southern California
Structural priming research shows that recent exposure to a syntactic structure facilitates later processing of that structure (e.g. Bock 1986). However, this research tacitly assumes (cf. Luka & Barsalou 2005), that comprehenders build in-depth, fully-detailed structures despite evidence that we often build shallow, underspecified representations (e.g. Christianson et al 2001). This suggests that priming is possible even in highly underspecified structures. We use ungrammatical, potentially 'unbuildable' sentences like Complex-NP-Constraint islands (CNPC) and Subject-islands (ex.1) to test extreme cases of underspecification.
(1a) CNPC-island: *Who did John deny the claim that the princess married?
(1b) Subject-island:*What did opponents of start a violent riot outside the mall?
We manipulated (between-subjects) prime-target proximity (Bock & Griffin 2000): primes and targets were CNCP-islands and Subject-islands separated by one or five unrelated sentences (Lag1/Lag5). Participants (n=84) rated acceptability on a 5-point scale.
Lag1 results showed recent exposure to a prime improves CNPC-island and Subject-island acceptability, but this effect is stronger with CNPC-islands. No priming was observed for Lag5. This suggests: (1) Short-term priming may be possible for some highly underspecified structures, but may be modulated by the type of structure. (2) Though underspecified, representations were fine-grained enough to distinguish between two types of grammaticality violations.