UC San Diego

This dissertation features a relative clause island, whose status has been known to differ significantly across languages and extraction types. By conducting a series of acceptability judgment experiments with human participants, as well as measuring token-by- token surprisal values among large language models, I demonstrate the following: First, sentences with relativization out of another relative clause (double relative) in Japanese, some of which have been claimed to be well-formed, still display a drop in acceptability indicative of the penalty of an island violation (Chapter 2). Second, the small penalty of relative clause island violation does not appear to be due to aggregating inter-item and/or inter-participant variability in acceptability judgments (Chapter 3). Third, the fact that most sentences with double relative in English display a much larger drop in acceptability can be accounted for by additional factors pertaining to the structure (Chapter 4). And lastly, the acceptability of Japanese scrambling out of a relative clause, which has been judged as ill-formed without exception, can be improved (Chapter 5). Altogether, I show that the penalty of violating a relative clause island is relatively small regardless of language and extraction type, and what people have perceived as a relative clause island effect should be deconstructed into the penalty per se and a group of additional factors that are processing, syntax, and semantics-oriented.