Processing Complex NP Islands in Hebrew

Since Ross's (1967) work on island constraints until many present day works, the treatment of island phenomena is mostly based on universal constraints on the grammar. However, it has been claimed that the unacceptability of certain island violations can be explained by extragrammatical factors, such as processing di culties. The advantage of this approach is that the unacceptability is explained by factors that exist independently of the island phenomena. The aim of this work is to investigate whether processing factors known to a ect ller-gap dependencies (or wh-movement) can account for the unacceptability of Complex NP island violations in Hebrew, including both extractions from complement clauses of NPs (CNPCC) and relative clauses of NPs (CNPRC). The e ect of D-linking, a manipulation over the wh-element that is independently associated with the processing of ller-gap dependencies, is examined in a series of experiments. Two acceptabilityrating experiments examine the e ect of D-linking on CNPCC and CNPRC violations, and a selfpaced reading experiment complements the results of the acceptability study on CNPCC. Experiment 1 shows that the D-linking of the wh-phrase improves the acceptability judgments of questions with CNPCC violations in a statistically signi cant manner, although this manipulation does not make the sentences fully acceptable. I suggest that questions with CNPCC violations and D-linked wh-phrases receive better judgments because the D-linked wh-phrases are more easily integrated into the sentence representation. However, these ndings cannot base the view that the low acceptability of CNPCC violations as a whole is due to processing limitations. Experiment 2 shows no signi cant e ect for D-linking in CNPRC islands, suggesting that CNPRC violations are not susceptible to processing factors. Experiment 3 reveals that D-linking facilitates the processing of the ller-gap dependencies in sentences with CNPCC violations. However, processing di erences within CNPCC violations and a grammatical baseline are not matched by contrasts in acceptability judgments, which is not compatible with the view that limitations on the cognitive resources related to the ller-gap dependencies process are responsible for CNPCC violation e ects. The experimental results strongly imply that CNPRC and CNPCC are two very distinct phenomena and I argue that the unacceptability of CNPRC violations is due to their ungrammaticality. Regarding CNPCC violations, part of their unacceptability is due to the processing di culties involved in parsing a sentence with a ller-gap dependency. These processing di culties seem not to be the reason for the unacceptability of the violation as a whole, and they seem to explain the same kind of acceptability variation that is found between regular yes-no questions and wh-questions. However, for CNPCCs, there is a major slowdown in processing at the head of the island. This suggests either that the parser discovers the grammatical violation at this region, which provokes both the processing slowdown and the unacceptability; or that the unacceptability is a consequence of the processing e orts of building the derivation of the island, that is, sentences with CNPCC violations are generated by the grammar but are hard to process.