[4Rin1] Interactive 2

[4Rin1-15] Japanese Temporal Relation Classification with joint Dependency and Contextual Representations

〇Chenjing Geng¹, Fei Cheng², Lis Kanashiro Pereira¹, Masayuki Asahara³, Ichiro Kobayashi¹ (1.Ochanomizu University, 2.Kyoto University, 3.National Institute for Japanese Language and Linguistics)

Keywords:natural language processing, temporal relation classification, neural network

Recently, quite a few number of studies have been progressive for temporal relation extraction from a corpus,which is an important work used in several natural language processing application. However, less concentration had paid to corpus of Japanese. In this work, we explored the feasibility of applying neural networks to temporal relation identification in the non-English data (BCCWJ-TimeBank). We explored the strength of combining contextual word representations (CWR) and shortest dependency paths (SDP) for Japanese temporal relation classification.We carefully designed a set of experiments to gradually reveal the improvements contributed by CWR and SDP.The empirical results suggested following conclusions: 1) SDP offers richer information for beating the experiments with only source and target words. 2) CWR significantly outperforms fasttext. 3) In most cases, CWR + SDP achieves the best performance overall.