Rits-QA: List Answer Detection and Context Task with Zero Anaphora Handling

In this paper, we describe RitsQA: Ritsumeikan question answering system. We have improved our QA system submitted in QAC1 in subcategorization of NE type and question type analysis and in handling numerical expressions using corpus based approach. We also participated subtask 2 and 3 using these core QA systems. We have implemented list answer detection module for subtask 2 and ellipsis handling module for subtask 3. We had evaluations of subtask 1, 2 and 3. According to Formal Run evaluation, our QA system had some progress in some aspects. Especially in subcategorization of NE types, we have obtained better improvement in performance of core QA system. In subtask 2 and 3, we have a little success in detection of the number of answers from surface pattern of question sentence. Also, in ellipsis handling, almost 37% of follow-up questions was correctly treated in ellipsis. However, it is necessary to expand patterns of ellipsis and improve list answer processing.