Subjacency as a processing phenomenon

Abstract

In this study we investigated the hypothesis that standard subjacency effects in so-called “wh-islands” are not necessarily due to an innate syntactic constraint, i.e. a problem of language competence, but rather to limits on the human Sentence processor, i.e. a problem of performance. We did so by gathering global acceptability judgements and by measuring event-related brain potentials (ERPs) in response to both yes/no- and wh-questions containing embedded that-, if- and wh-clauses. The embedding of any one of these clause types within a yes/no-question typically results in a well-formed sentence. The well-formedness of wh-questions, on the other hand, depends in large part on the type of embedded clause into which a syntactic dependency is formed: dependencies into embedded that-clauses are usually considered grammatical, while dependencies into embedded if-clauses are considered marginal and dependencies into wh-clauses (“wh-islands”) ungrammatical. We predicted that these differences in grammaticality across wh-question types could be derived from an interaction of (a) the lexical semantic processing effects induced by the choice of that, if or who/what at the embedded clause boundary with (b) the syntactic processing effects of maintaining a dependency between a main clause wh-phrase (“filler”) and an embedded clause position (“gap”) across an embedded clause boundary. In our acceptability judgement studies (Experiments 1 and 2), we correspondingly found main effects of embedded clause type, main effects of question type and an interaction of the two. In our ERP study (Experiment 3), lexicosemantic processing effects were indexed by differential N400 amplitude at the embedded clause boundary: the N400 response to the interrogative pronouns who and what was greater than the N400 response to the complementiser if, which was in turn greater than the N400 response to the complementiser that. Syntactic processing effects of holding a wh-filler in working memory pending assignment to a gap, or of retrieving it from memory for purposes of gap assignment, were indexed by a negative component between 300 and 500 msec that was largest over left anterior regions. An interaction of the two effects was seen in the assignment of main clause fillers to embedded clause gaps; while gap assignment at this position in wh-questions was also indexed by left anterior negativity, this negativity was modulated in the same way as the lexically induced N400 elicited at the clause boundary: largest in wh-clauses, smallest in that-clauses and intermediate in size in if-clauses. Similar effects were seen in the ERPs to sentence-final words of wh-questions. We take these findings as evidence that subjacency violations in wh-island contexts may be attributed to the interaction of lexical semantic processing factors at the embedded clause boundary with the necessity of holding a filler in working memory.