Constraint Use In Language Comprehension

本研究旨在探讨正常成人语言理解中句法歧义的消解。对歧义消解的研究可以阐明语言理解系统的结构，也可以评价有关语言心理表征本质的心理语言学和语言学理论。该研究的理论框架是一个词汇的“基于约束”的框架，在这个框架中，句法歧义被视为词汇歧义的一种，因此句法歧义的解决与词汇语义歧义的解决具有相同的性质。这种方法已被有效地应用于英语中几种“经典”语法歧义的分析，这些歧义以前被认为是一种完全不同的处理体系结构的证据。实验采用自定速阅读法和眼注视监测来研究歧义解决和约束使用。还收集了大量关于约束强度的规范性数据，并使用回归方法来确定不同约束在不同时间点对理解者解释的影响程度。这项研究的重要之处在于，它提出了一种统一的语言理解解释，其中所有层次的表征（语音、词汇、句法、话语等）都受到相同的基本属性的支配，包括频率敏感性和通过约束满足激活表征。语言表征的统一解释不仅影响人类语言使用的理论，也影响人工智能的工作，迄今为止，人工智能通常反映了更传统的人类语言理解理论。因此，我们对人类歧义解决的统一方法的理解可能对计算机自然语言理解研究产生重大影响，特别是在这些系统中非常困难的歧义解决问题。

This research is designed to investigate syntactic ambiguity resolution in normal, adult language comprehension. The study of ambiguity resolution can illuminate the architecture of the language comprehension system and can evaluate the psycholinguistic and linguistic theories concerning the nature of linguistic mental representations. The theoretical framework of the investigation is a lexical, "constraint-based" framework in which syntactic ambiguity is seen as a type of lexical ambiguity, such that syntactic ambiguity resolution has the same properties as the resolution of lexical semantic ambiguities. This approach has been applied profitably to the analysis of several "classic" syntactic ambiguities in English that had previously been taken as evidence for a radically different processing architecture. The experiments use self-paced reading methods and eye fixation monitoring to study ambiguity resolution and constraint use. A large amount of normative data is also collected concerning the strength of the constraints, and regression methods are used to determine the extent to which difference constraints affect comprehenders' interpretations at different points in time. The research is significant in that it advances a unified account of language comprehension in which representation at all levels (phonological, lexical, syntactic, discourse, etc.) are governed by the same basic properties, including frequency sensitivity and activation of representations via constraint satisfaction. A unified account of language representation affects not only theories of human language use, but also work in artificial intelligence, which to date has generally reflected the more traditional theories of human language comprehension. Our understanding in the unified approach of human ambiguity resolution could therefore have substantial impact on computer natural language comprehension research, particularly the very difficult problem of ambiguity resolut ion in these systems.