Chinese Tense Labelling and Causal Analysis

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 2227–2237, Osaka, Japan, December 11-17 2016. Chinese Tense Labelling and Causal Analysis Hen-Hsen Huang, Chang-Rui Yang, and Hsin-Hsi Chen Department of Computer Science and Information Engineering National Taiwan University No. 1, Sec. 4, Roosevelt Road, Taipei, 10617 Taiwan {hhhuang, cjyang}@nlg.csie.ntu.edu.tw; hhchen@ntu.edu.tw Abstract This paper explores the role of tense information in Chinese causal analysis. Both tasks of causal type classification and causal directionality identification are experimented to show the significant improvement gained from tense features. To automatically extract the tense features, a Chinese tense predictor is proposed. Based on large amount of parallel data, our semi- supervised approach improves the dependency-based convolutional neural network (DCNN) models for Chinese tense labelling and thus the causal analysis. 1 Introduction Causal analysis plays a crucial role in the applications such as event extraction (Hashimoto et al., 2012; 2014), causality inference (Tanaka et al., 2012), question-answering (Oh et al., 2013), and motivation identification (Nguyen et al., 2015). Compared to English, the topic of causal analysis in Chinese is rarely touched. In this work, we explore the role of tense information in Chinese causal analysis. As pointed by Mirza (2014), the causal relation and temporal information is correlated. In a causal rela- tion, the cause intuitively precedes its effect. In other words, the tense information could be useful fea- tures in the tasks of causal analysis. Two tasks of causal analysis are investigated in this study: causal type classification and causal di- rectionality identification. The Chinese discourse relation corpus, Chinese Discourse Treebank (CDTB) (Li et al., 2014), is adopted as our dataset. In CDTB, six types of causality relations, Purpose, Back- ground, Hypothetical, Inference, Condition, and Cause-Result, are defined. A discourse relation connects two arguments. In the case of causality, one of the two arguments (e.g., arg1) presents a situation, and it is causally affected by the other argument (e.g., arg2). For ex- ample, the first part of the sentence (S1) shows a reason, and the second part, which is underlined, is its effect. (S1) 由於產能不足，國內自給率不到四成，大部分要仰賴進口。 (Because of in- sufficient capacity, the domestic self-sufficiency rate is less than 40, most rely on imports.) The direction of arg1 and arg2 is reversible. Like (S1), the reason is described in the former argu- ment in most cases, and the effect is presented in the latter argument. However, (S2) shows a counter- example that presents the effect in the former part. To exactly extract the cause and the effect in natu- ral language, causal directionality identification is required. (S2) 西藏銀行部門積極調整信貸結構，以確保農牧業生產等重點產業的投入，加大對 工業，能源，交通，通信等建設的正常資金供應量。(Tibet banking sector ac- tively adjust credit structure in order to ensure the input of agricultural production and other key industries, and increase the industrial, energy, transportation, communications, con- struction of the normal supply of funds.) There is no tense annotation in the CDTB. For this reason, we select all the samples of causality re- This work is licensed under a Creative Commons Attribution 4.0 International Licence. Licence details: http://creativecommons.org/licenses/by/4.0/ 2227 lation from CDTB, …

Content truncated to 500 words