Discovering Dialog Structure Graph for Coherent Dialog Generation

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 1726–1739 August 1–6, 2021. ©2021 Association for Computational Linguistics 1726 Discovering Dialog Structure Graph for Coherent Dialog Generation Jun Xu1∗, Zeyang Lei2∗, Haifeng Wang2, Zheng-Yu Niu2, Hua Wu2, Wanxiang Che1† 1Research Center for Social Computing and Information Retrieval, Harbin Institute of Technology, Harbin, China 2Baidu Inc., Beijing, China {jxu, car}@ir.hit.edu.cn, {leizeyang, wanghaifeng, niuzhengyu, wu hua}@baidu.com Abstract Learning discrete dialog structure graph from human-human dialogs yields basic insights into the structure of conversation, and also pro- vides background knowledge to facilitate dia- log generation. However, this problem is less studied in open-domain dialogue. In this pa- per, we conduct unsupervised discovery of discrete dialog structure from chitchat cor- pora, and then leverage it to facilitate coher- ent dialog generation in downstream systems. To this end, we present an unsupervised model, Discrete Variational Auto-Encoder with Graph Neural Network (DVAE-GNN), to discover discrete hierarchical latent dialog states (at the level of both session and utterance) and their transitions from corpus as a dialog structure graph. Then we leverage it as background knowledge to facilitate dialog management in a RL based dialog system. Experimental re- sults on two benchmark corpora confirm that DVAE-GNN can discover meaningful dialog structure graph, and the use of dialog structure as background knowledge can significantly im- prove multi-turn coherence. 1 Introduction With the aim of building a machine to converse with humans naturally, some work investigate neu- ral generative models (Shang et al., 2015; Serban et al., 2017). While these models can generate lo- cally relevant dialogs, they struggle to organize in- dividual utterances into globally coherent flow (Yu et al., 2016; Xu et al., 2020b). The possible reason is that it is difficult to control the overall dialog flow without background knowledge about dialog structure.1 However, due to the complexity of open- domain conversation, it is laborious and costly to annotate dialog structure manually. Therefore, it is ∗Equal contribution. †Corresponding author: Wanxiang Che. 1Dialog structure means dialog states and their transitions. of great importance to discover open-domain dia- log structure from corpus in an unsupervised way for coherent dialog generation. Some studies tried to discover dialog structure from task-oriented dialogs (Shi et al., 2019). How- ever, the number of their dialog states is limited to only dozens or hundreds, which cannot cover fine- grained semantics in open-domain dialogs. Fur- thermore, the dialog structures they discovered generally only contain utterance-level semantics (non-hierarchical), without session-level semantics (chatting topics) that are essential in open-domain dialogs (Wu et al., 2019; Kang et al., 2019; Xu et al., 2020c).2 Thus, in order to provide a full pic- ture of open-domain dialog structure, it is desirable to discover a two-layer directed graph that contains session-level semantics in the upper-layer vertices, utterance-level semantics in the lower-layer ver- tices, and edges among these vertices. In this paper, we propose a novel discrete vari- ational auto-encoder with graph neural network (DVAE-GNN) to discover a …

Content truncated to 500 words