Linguistic knowledge based discourse modeling for context question answering

This dissertation is motivated by the recent developments in scenario-based context question answering (QA). The role of discourse processing, and its implication on query expansion for a sequence of questions, is investigated. My view is that a question sequence is not random, but rather, follows a coherent manner to serve certain information goals of users. Therefore, this sequence of questions can be considered as a mini discourse with some characteristics of discourse cohesion and discourse coherence. Understanding such a discourse will help QA systems better interpret questions and retrieve answers. Thus, in the first part of my study, I propose three models driven by Centering Theory for discourse processing: an anaphora model that resolves pronoun references for each question, a forward model that makes use of the forward looking centers from previous questions, and a transition model that takes into account the transition state between adjacent questions. The empirical results indicate that more sophisticated processing based on discourse transitions and centers can significantly improve the performance of document retrieval compared to models that only resolve references. In the second part of the study, the influence of the processing based on pronoun resolution and definite description resolution is investigated. Results show that a combined model that incorporates both approaches performs the best under the situation where no explicit target is given for the context questions. The processing for the event type of context question answering is also investigated briefly. For different discourse models proposed in the dissertation, systematic evaluation is provided and the potentials and limitations of these models in processing coherent context questions are also discussed.