A Joint Model for Dropped Pronoun Recovery and Conversational Discourse Parsing in Chinese Conversational Speech

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 1752–1763 August 1–6, 2021. ©2021 Association for Computational Linguistics 1752 A Joint Model for Dropped Pronoun Recovery and Conversational Discourse Parsing in Chinese Conversational Speech Jingxuan Yang1, Kerui Xu1, Jun Xu2,3∗, Si Li1, Sheng Gao1, Jun Guo1, Nianwen Xue4 and Ji-Rong Wen2,3 1School of Artificial Intelligence, Beijing University of Posts and Telecommunications 2Gaoling School of Artificial Intelligence, Renmin University of China 3Beijing Key Laboratory of Big Data Management and Analysis Methods 4Department of Computer Science, Brandeis University {yjx, xukerui, lisi, gaosheng, guojun}@bupt.edu.cn junxu@ruc.edu.cn, jirong.wen@gmail.com, xuen@brandeis.edu Abstract In this paper, we present a neural model for joint dropped pronoun recovery (DPR) and conversational discourse parsing (CDP) in Chi- nese conversational speech. We show that DPR and CDP are closely related, and a joint model benefits both tasks. We refer to our model as DiscProReco, and it first encodes the tokens in each utterance in a conversa- tion with a directed Graph Convolutional Net- work (GCN). The token states for an utter- ance are then aggregated to produce a sin- gle state for each utterance. The utterance states are then fed into a biaffine classifier to construct a conversational discourse graph. A second (multi-relational) GCN is then ap- plied to the utterance states to produce a dis- course relation-augmented representation for the utterances, which are then fused together with token states in each utterance as input to a dropped pronoun recovery layer. The joint model is trained and evaluated on a new Structure Parsing-enhanced Dropped Pronoun Recovery (SPDPR) dataset that we annotated with both two types of information. Experi- mental results on the SPDPR dataset and other benchmarks show that DiscProReco signifi- cantly outperforms the state-of-the-art base- lines of both tasks. 1 Introduction Pronouns are often dropped in Chinese conversa- tions as the identity of the pronoun can be inferred from the context (Kim, 2000; Yang et al., 2015) without causing the sentence to be incomprehensi- ble. The task of dropped pronoun recovery (DPR) aims to locate the position of the dropped pronoun and identify its type. Conversational discourse pars- ing (CDP) is another important task that aims to analyze the discourse relations among utterances ∗ Corresponding author A1: 张 老师 在家 吗 ？ Is Miss Zhang at home ? B1: 她 现在 不 在家 ，今天 她 在 外面 上课 。 She is not home right now. She is teaching outside. A2: 我 可以 给 她 打 个 电话 吗 ？ Can I give her a call ? B2: 下午 四点 以后 她 应该 有 时间 接 电话 。 She should have time to answer the phone after 4 p.m. B4: 或者 你 明天 早上 再 打 Or you can call her tomorrow morning. B5: 明天 她 在家 。 She will be at home tomorrow. question reply expansion expansion reply reply B3: (你) 可以 给 她 打 个 电话 (You) can give her a call. (Pro-drop utterance) pleonastic Figure …

Content truncated to 500 words