Reading Turn by Turn: Hierarchical Attention Architecture for Spoken Dialogue Comprehension

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 5460–5466 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 5460 Reading Turn by Turn: Hierarchical Attention Architecture for Spoken Dialogue Comprehension Zhengyuan Liu, Nancy F. Chen Institute for Infocomm Research, A*STAR {liu zhengyuan, nfychen}@i2r.a-star.edu.sg Abstract Comprehending multi-turn spoken conversa- tions is an emerging research area, present- ing challenges different from reading com- prehension of passages due to the interactive nature of information exchange from at least two speakers. Unlike passages, where sen- tences are often the default semantic model- ing unit, in multi-turn conversations, a turn is a topically coherent unit embodied with imme- diately relevant context, making it a linguis- tically intuitive segment for computationally modeling verbal interactions. Therefore, in this work, we propose a hierarchical attention neural network architecture, combining turn- level and word-level attention mechanisms, to improve spoken dialogue comprehension per- formance. Experiments are conducted on a multi-turn conversation dataset, where nurses inquire and discuss symptom information with patients. We empirically show that the pro- posed approach outperforms standard atten- tion baselines, achieves more efficient learning outcomes, and is more robust to lengthy and out-of-distribution test samples. 1 Introduction Reading comprehension has attracted much inter- est in the past couple years, fueled by avid neural modeling investigations. Given a certain textual content, the goal is to answer a series of questions based on implicit semantic understanding. Previ- ous work has focused on passages like Wikipedia (Rajpurkar et al., 2016) or news articles (Hermann et al., 2015). Recently, dialogue comprehension in the form of cloze tests and multi-choice questions has also started to spur research interest (Ma et al., 2018; Sun et al., 2019). Different from passages, human-to-human dialogues are a dynamic and in- teractive flow of information exchange, which are often informal, verbose and repetitive.1 This leads to lower information density and more topic dif- fusion, since the spoken content of a conversation is determined by two speakers, each with his/her own thought process and potentially distracting and parallel streams of thoughts. To address such challenges, we propose to uti- lize a hierarchical attention mechanism for di- alogue comprehension, which has shown to be effective in various natural language processing tasks (Yang et al., 2016; Choi et al., 2017; Hsu et al., 2018). The hierarchical models successively capture contextual information at different levels of granularity, leveraging coarse-grained attention to reduce the potential distraction in finer-grained attention but at the same time exploit finer-grained attention to distill key information for downstream tasks more precisely and efficiently. While in document tasks sentences are the de- fault semantic modeling unit at the coarse-grained level, utterances might be a more suitable counter- part in spoken dialogues, as dialogues often con- sist of incomplete sentences. However, a single utterance/sentence which usually implies informa- tion from one speaker is insufficient for grasping the full relevant context, as the interactive infor- mation from the interlocutor is often necessary. In multi-turn dialogues, each turn is one round …

Content truncated to 500 words