STRUDEL: Structured Dialogue Summarization for Dialogue Comprehension

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 4949 - 4958 December 7-11, 2022 ©2022 Association for Computational Linguistics STRUDEL : Structured Dialogue Summarization for Dialogue Comprehension Borui Wang1 Chengcheng Feng1 Arjun Nair1 Madelyn Mao1 Jai Desai1 Asli Celikyilmaz2 Haoran Li2 Yashar Mehdad2 Dragomir Radev1 1Yale University 2Meta AI {borui.wang, dragomir.radev}@yale.edu {aslic, aimeeli, mehdad}@fb.com Abstract Abstractive dialogue summarization has long been viewed as an important standalone task in natural language processing, but no previous work has explored the possibility of whether abstractive dialogue summarization can also be used as a means to boost an NLP system’s performance on other important dialogue com- prehension tasks. In this paper, we pro- pose a novel type of dialogue summarization task - STRUctured DiaLoguE Summarization (STRUDEL ) - that can help pre-trained lan- guage models to better understand dialogues and improve their performance on important dialogue comprehension tasks. In contrast to the holistic approach taken by the traditional free-form abstractive summarization task for dialogues, STRUDEL aims to decompose and imitate the hierarchical, systematic and struc- tured mental process that we human beings usually go through when understanding and analyzing dialogues, and thus has the advan- tage of being more focused, specific and in- structive for dialogue comprehension models to learn from. We further introduce a new STRUDEL dialogue comprehension modeling framework that integrates STRUDEL into a di- alogue reasoning module over transformer en- coder language models to improve their dia- logue comprehension ability. In our empirical experiments on two important downstream di- alogue comprehension tasks - dialogue ques- tion answering and dialogue response predic- tion - we demonstrate that our STRUDEL dia- logue comprehension models can significantly improve the dialogue comprehension perfor- mance of transformer encoder language mod- els. 1 Introduction In natural language processing, abstractive dia- logue summarization (Feng et al., 2021) has long been viewed as an important standalone task, but no previous work has explored the possibility of whether abstractive dialogue summarization can Dialogue Pre-Trained Language Model STRUDEL Question Answering (Other Tasks) Upstream Task: Structured Dialogue Summarization Downstream Tasks: Dialogue Comprehension Meta-Model Response Prediction … Figure 1: STRUDEL as a meta-model on top of pre- trained language models for dialogue comprehension. also be used as a means to boost an NLP system’s performance on other important dialogue compre- hension tasks. When performing language under- standing, a very natural and effective first step that human beings usually take in their mental process is to try to summarize the main content of a piece of text, usually from multiple perspectives each focus- ing a different aspect of the text. This is especially true when human readers or speakers are trying to understand a dialogue or a conversation, which involve multi-turn exchange of information follow- ing a general theme, topic or storyline. Therefore, we would like to ask the following question - can the task of abstractive dialogue summarization also help NLP models to learn to perform better dia- logue comprehension? In this paper, we propose a novel type …

Content truncated to 500 words