CoLV: A Collaborative Latent Variable Model for Knowledge-Grounded Dialogue Generation

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 2250–2261 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 2250 CoLV: A Collaborative Latent Variable Model for Knowledge-Grounded Dialogue Generation Haolan Zhan1∗, Lei Shen1,2 ∗, Hongshen Chen3†, and Hainan Zhang3 1 University of Chinese Academy of Sciences, Beijing, China 2Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China 3Data Science Lab, JD.com, Beijing, China zhanhaolan316@gmail.com, shenlei17z@ict.ac.cn, ac@chenhongshen.com Abstract Knowledge-grounded dialogue generation has achieved promising performance with the en- gagement of external knowledge sources. Typ- ical approaches towards this task usually per- form relatively independent two sub-tasks, i.e., knowledge selection and knowledge-aware re- sponse generation. In this paper, in order to improve the diversity of both knowledge se- lection and knowledge-aware response gener- ation, we propose a collaborative latent vari- able (CoLV) model to integrate these two as- pects simultaneously in separate yet collabo- rative latent spaces, so as to capture the in- herent correlation between knowledge selec- tion and response generation. During genera- tion, our proposed model firstly draws knowl- edge candidate from the latent space condi- tioned on the dialogue context, and then sam- ples a response from another collaborative la- tent space conditioned on both the context and the selected knowledge. Experimental results on two widely-used knowledge-grounded di- alogue datasets show that our model outper- forms previous methods on both knowledge se- lection and response generation. 1 Introduction Knowledge-grounded dialogue generation (Liu et al., 2018; Zhou et al., 2018a; Lian et al., 2019; Tian et al., 2020), which utilizes external knowl- edge to enhance conversation backgrounds, has achieved promising performance. To exploit ex- ternal knowledge efficiently for conversations, typ- ical approaches (Dinan et al., 2019; Kim et al., 2020; Xu et al., 2020; Sun et al., 2020; Chen et al., 2020; Meng et al., 2020; Chen et al., 2021) tend to decompose this task into two streamlined sub-tasks: knowledge selection and knowledge- aware response generation. Besides, some other work (Qin et al., 2019; Tian et al., 2020) also tries to ∗First two authors contribute equally. †Corresponding author. Dialogue context What is your favorite number? → I love the number 7. What do you think about that? 1. Anyone who dares to kill Cain "will suffer vengeance seven times over". 2. Seven is the natural number following six and preceding eight. Knowledge candidates 3. Islam first came to the western coast when Arab traders as early as the 7th century CE. 4. The number 7 has been associated with a great deal of symbolism in religion. In western culture, it is often con- sidered lucky. ...... N. This genre has been popular throughout the history of culture. Response a Yeah. I know that it is before 8 and after 6! Response b Yes, it is known as a lucky number in western countries! Response c I think 7 is lucky certain cultures. It also depicts some reli- gious importance. Table 1: An example of knowledge-grounded conver- sations. Given the dialogue context, knowledge selec- tion and …

Content truncated to 500 words