This paper revisits the key insight of Ross (1967) that while embedded interrogative (+Q) sentences are islands for wh-extraction, embedded declarative (-Q) sentences are not. Following Bresnan (1970), we observe that verbs which embed declaratives have predicable complements (that or Ø), whereas verbs which embed interrogatives subcategorize for a wide array of +Q continuations (whether, if, and embedded questions). We hypothesize that the incremental ambiguity of wonder class verbs which embed +Q elements results in greater cognitive load for a predictive, incremental parser, triggering weak islandhood. We model results from Alexopoulou & Keller (2007), who found Magnitude Estimates for wh-extraction from sententials embedded by wonder to be degraded compared to controls with embedding verb claim. We construct and weight a Derivational Minimalist Grammar (Stabler, 1997) which renders the Bresnan (1970) facts. We model the ideal comprehender's performance on the Alexopoulou & Keller (2007) sentences using Entropy Reduction (Hale, 2006). On our model, Island condition sentences conveyed an average total of 3.379 bits of information, compared to Nonisland sentences, which conveyed 1.899 bits of information.
Proceedings of the 29th West Coast Conference on Formal Linguistics
edited by Jaehoon Choi, E. Alan Hogue, Jeffrey Punske, Deniz Tat, Jessamyn Schertz, and Alex Trueman
Table of contents