A Distance-Aware Multi-Task Framework for Conversational Discourse Parsing

Proceedings of the 29th International Conference on Computational Linguistics, pages 912–921 October 12–17, 2022. 912 A Distance-Aware Multi-Task Framework for Conversational Discourse Parsing Yaxin Fan, Peifeng Li, Fang Kong, and Qiaoming Zhu∗ School of Computer Science and Technology, Soochow University, Suzhou, China yxfansuda@stu.suda.edu.cn {pfli, kongfang, qmzhu}@suda.edu.cn Abstract Conversational discourse parsing aims to con- struct an implicit utterance dependency tree to reflect the turn-taking in a multi-party conver- sation. Existing works are generally divided into two lines: graph-based and transition- based paradigms, which perform well for short- distance and long-distance dependency links, respectively. However, there is no study to consider the advantages of both paradigms to facilitate conversational discourse parsing. As a result, we propose a distance-aware multi- task framework DAMT that incorporates the strengths of transition-based paradigm to facili- tate the graph-based paradigm from the encod- ing and decoding process. To promote multi- task learning on two paradigms, we first intro- duce an Encoding Interactive Module (EIM) to enhance the flow of semantic information between both two paradigms during the encod- ing step. And then we apply a Distance-Aware Graph Convolutional Network (DAGCN) in the decoding process, which can incorporate the different-distance dependency links predicted by the transition-based paradigm to facilitate the decoding of the graph-based paradigm. The experimental results on the datasets STAC and Molweni show that our method can signifi- cantly improve the performance of the SOTA graph-based paradigm on long-distance de- pendency links. Our code is available at https://github.com/yxfanSuda/DAMT. 1 Introduction The goal of conversational discourse parsing is to uncover latent conversation topics and construct an implicit utterance dependency tree to reflect the turn-taking in a multi-party conversation. Since the discourse structure is essential to understand multi- party conversations, it has been widely applied to various Natural Language Processing (NLP) ap- plications,such as response generation (Hu et al., ∗Corresponding author (u1) A: headers , kernel image and all the rest (u2) B: what project are you trying to build ? (u3) A: this did n't work either , i guess that the k cc is correctly registered but fails . (u4) C: his gcc segfaults on helloworld (u5) B: ahh , so , a libc issue most likelyis there a kk pastebin url of the errors ? (u6) D: are you using any backports or sth ? (u7) B: can you pastebin the results of that `` gcc kk -v test.c '' if there are errors ? (u8) B: do you have the matching kernel-headers kk and glibc-devel packages installed ? (u9) A: can you help me how to find that out ? 701 Figure 1: A multi-party dialogue with its dependency structure, where the solid lines, dotted lines and dashed lines denote the relations “Clarification Question”, “Comment”, and “Result” respectively, and A, B, C, D mean different speakers. 2019), reading comprehension(Li et al., 2021b; Li and Zhao, 2021), meeting summarization (Feng et al., 2021), and emotion recognition(Sun et al., 2021). Segmented Discourse Representation Theory (SDRT) (Asher et al., 2003) is one of the most influential theories …

Content truncated to 500 words