Establishment of Dependency between Head and Arguments in Japanese Scrambling Constructions

本研究探讨了不同类型的打乱成分对加工打乱结构所需工作记忆资源的影响，以及打乱成分在加工日语打乱结构时激活水平的变化。首先，构建阅读广度和听力广度测试。其次，当一个关系从句的头NP有两个潜在的连接位点，即[RC] NP- gen NP时，研究了关系从句的连接偏好。复合NP是主句的对象，它要么被置乱，要么原地不动。采用离线可接受性判断范式和时间可接受性判断范式考察依恋偏好。附件偏好根据处理混乱和规范词序所需的不同WM资源而变化。在日语中，在关系从句结构中，间隔出现在相应先行词出现之前。研究结果表明，不同词序的词源对语篇间隙及其先行词的建立有一定的影响。结果还表明，某些类型的打乱句子比规范句子需要更少的处理成本。第三，使用离线任务对二语进行的进一步研究表明，二语熟练程度对解决歧义关系从句附件也有影响。最后，通过使用跨模态启动范式，在涉及填补空白效应的句子中检查了混乱参数NPs（例如间接宾语）的激活水平。结果表明，解析器在假间隙位置激活了一个打乱的成分（填充），但是当重新分析发生时，曾经激活的填充被抑制，并在真正的间隙位置重新激活。同时还观察了个体WM能力差异的影响。上述结果表明，根据结构和可用工作记忆资源的不同，打乱后的论点与它们的头脑之间形成了依赖关系。少

The present study investigated how different types of scrambled constituent influenced processing scrambling constructions with respect to the amount of required working memory resources and how activation levels of scrambled constituents varied while Japanese scrambling constructions were processed.Firstly, reading-span and listening-span tests were constructed. Secondly, relative-clause attachment preferences were examined when the head NP of a relative clause had two potential attachment sites, i. e., [RC] NP-Gen NP. The complex NP was the object of the main clause and either underwent scrambling or remained in-situ. Attachment preferences were examined by using offline and timed acceptability judgment paradigms. Attachment preferences varied according to different WM resources that were required to process scrambled and canonical word orders. In Japanese gaps appear before the appearance of corresponding antecedents in relative-clause constructions. The results indicated that WM re … More sources of different word orders influenced the establishment of gap and its antecedent. The results also indicated the possibility that some types of scrambled sentences required less processing cost than canonical sentences. Thirdly, further researches on L2 using off-line tasks indicated that in resolving ambiguous relative-clause attachments in L2 proficiency levels were also involved.Finally, activation levels of scrambled argument NPs (e. g., indirect object) were examined in sentences that involved a filled-gap effect by using the cross-modal priming paradigm. The results indicated that the parser activated a scrambled constituent (filler) at false gap sites but when a reanalysis occurred the once activated filler was suppressed and it was reactivated at the true gap site. The influence of individuals' difference of WM capacity was also observed.The above-mentioned results indicated that dependencies were formed between scrambled arguments and their heads by re-and de-activating scrambled arguments repeatedly according to the structures and available working memory resources. Less