Adversarial Connective-exploiting Networks for Implicit Discourse Relation Classification

Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 1006–1017 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1093 Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 1006–1017 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1093 Adversarial Connective-exploiting Networks for Implicit Discourse Relation Classification Lianhui Qin1,2, Zhisong Zhang1,2, Hai Zhao1,2,∗, Zhiting Hu3, Eric P. Xing3 1Department of Computer Science and Engineering, Shanghai Jiao Tong University 2Key Laboratory of Shanghai Education Commission for Intelligent Interaction and Cognitive Engineering, Shanghai Jiao Tong University, Shanghai, 200240, China 3Carnegie Mellon University {qinlianhui, zzs2011}@sjtu.edu.cn, zhaohai@cs.sjtu.edu.cn, {zhitingh, epxing}@cs.cmu.edu Abstract Implicit discourse relation classification is of great challenge due to the lack of con- nectives as strong linguistic cues, which motivates the use of annotated implicit connectives to improve the recognition. We propose a feature imitation frame- work in which an implicit relation net- work is driven to learn from another neu- ral network with access to connectives, and thus encouraged to extract similarly salient features for accurate classification. We develop an adversarial model to en- able an adaptive imitation scheme through competition between the implicit network and a rival feature discriminator. Our method effectively transfers discriminabil- ity of connectives to the implicit features, and achieves state-of-the-art performance on the PDTB benchmark. 1 Introduction Discourse relations connect linguistic units such as clauses and sentences to form coherent seman- tics. Identification of discourse relations can ben- efit a variety of downstream applications includ- ing question answering (Liakata et al., 2013), ma- chine translation (Li et al., 2014), text summariza- tion (Gerani et al., 2014), opinion spam detection (Chen and Zhao, 2015), and so forth. ∗Corresponding authors. This paper was partially sup- ported by Cai Yuanpei Program (CSC No. 201304490199 and No. 201304490171), National Natural Science Foun- dation of China (No. 61170114, No. 61672343 and No. 61272248), National Basic Research Program of China (No. 2013CB329401), Major Basic Research Program of Shang- hai Science and Technology Committee (No. 15JC1400103), Art and Science Interdisciplinary Funds of Shanghai Jiao Tong University (No. 14JCRZ04), and Key Project of National Society Science Foundation of China (No. 15- ZDA041). Connectives (e.g., but, so, etc) are one of the most critical linguistic cues for identifying dis- course relations. When explicit connectives are present in the text, a simple frequency-based map- ping is sufficient to achieve over 85% classifica- tion accuracy (Xue et al., 2016; Li et al., 2016). In contrast, implicit discourse relation recognition has long been seen as a challenging problem, with the best accuracy so far still lower than 50% (Chen et al., 2015). In the implicit case, discourse rela- tions are not lexicalized by connectives, but to be inferred from relevant sentences (i.e., arguments). For example, the following two adjacent sentences Arg1 and Arg2 imply relation Cause (i.e., Arg2 is the cause of Arg1). [Arg1]: Never mind. [Arg2]: You already know the answer. [Implicit connective]: Because [Discourse relation]: Cause Various attempts have been …

Content truncated to 500 words