Legal Question Answering System Using FrameNet.

A central issue of yes/no question answering is the usage of knowledge source given a question. While yes/no question answering has been studied for a long time, legal yes/no question answering largely differs from other domains. The most distinguishing characteristic is that legal issues require precise analysis of predicate argument structures and semantical abstraction in these sentences. We have developed a yes/no question answering system for answering questions for a statute legal domain. Our system uses a semantic database based on FrameNet, which works with a predicate argument structure analyzer, in order to recognize semantic correspondences rather than surface strings between given problem sentences and knowledge source sentences. We applied our system to the COLIEE (Competition on Legal Information Extraction/Entailment) 2018 task. Our frame based system achieved better scores on average than our previous system in COLIEE 2017, and was the second best score among participants of Task 4. We confirmed effectiveness of the frame information with the COLIEE training dataset. Our result shows the importance of the points described above, revealing opportunities to continue further work on improving our system's accuracy.