A Study On The Syntax-based Pronoun Resolution

This dissertation examines the issue of anaphora resolution, specifically pronoun resolution, from a syntax-based perspective. It aims to enhance the linguistic basis for pronoun resolution in Natural Language Processing, where Machine Translation, Information Extraction and Automatic Abstraction are requiring a sheer amount of linguistic knowledge to improve their performance. Different from previous research, this study emphasizes the importance of genre specific antecedent indicators in pronoun resolution.Five parts included, the dissertation starts with a brief introduction to anaphora resolution and the scope of research. Chapter two reviews the relevant notions, terminologies and theories. Chapter three is devoted to Mitkov's knowledge-poor approach, which selects antecedent by scoring NP candidates. Based on the observation that some antecedent indicators are genre specific, this study proposes, in Chapter four, to adjust and modify them for pronoun resolution in newspaper texts. The adjustment and modification can be regarded as an extension of Mitkov's approach for the newspaper genre.As a result of the adjustment and modification, the indicators are reduced, from Mitkov's original ten to the current six, with an optional indicator included, and the success rate for pronoun resolution in newspaper texts is raised from 54.9% to 88.2%. Further efforts can be made to get the parameters for different types of anaphora resolution in various genres.