The Question Answering System of DCUMT in NTCIR-11 QALab

This paper describes the question answering system developed at Dublin City University for participation in the QA Lab shared task in NTCIR-11 [20]. We participated in three tasks: center exam (multiple choice) tasks in Phases 1 and 2, and secondary exam task (written) in Phase 2. We built a QA system in which we use the specialized-purpose parser KBarse to acquire meaning representation which is called case frame graphs from history textbooks using commonsense knowledge. We used distributed representation for Out-Of-Vocaburary (OOV) words and missing assertions. We added prototype functionality for handling implicit arguments/relations, causality analysis, time analysis, and temporal order analysis by heuristics. Our results for center exam task in Phase 1 was 77.0 which was the first among seven submissions, for center exam task in Phase 2 was 72.0 which was the first among nine submissions, and for secondary exam task in Phase 2 in terms of precision were 71.4 (UTokyo), 62.5 (KyotoU), 71.8 (Hokkaido), 62.7 (Waseda), and 80.0 (Chuo) which were the first among two submissions.