Proceedings of the Annual Meeting of the Cognitive Science Society

The unacceptability of wh-extraction (e.g., question formation) out of certain syntactic structures, known as ‘island’ effects, has been a central topic in theoretical syntax for many years (Ross, 1967; Chomsky, 1973). A prominent example of islands is that extraction out of a sentential complement introduced by factive and manner-of-speaking verbs (‘What did John know/whisper that Mary bought?’) is less acceptable than extraction from a clause introduced by “bridge” verbs (‘What did John say that Mary bought?’). We aimed to replicate Ambridge and Goldberg (2008) who argued that extraction from a sentential complement is unacceptable in proportion to its discourse salience. We failed to replicate their results and found that there is no true island effect for such structures: instead there are separate, additive penalties based on two factors: (a) verb-frame frequency (cf. Dabrowska, 2008), and (b) the presence of extraction. These penalties give rise to apparent island effects as a result of the nonlinear relationship between true acceptability and acceptability ratings as measured in Likert scales and forced- choice tasks.