Regularizing Dialogue Generation by Imitating Implicit Scenarios

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 6592–6604, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 6592 Regularizing Dialogue Generation by Imitating Implicit Scenarios Shaoxiong Feng1 Xuancheng Ren2 Hongshen Chen3 Bin Sun1 Kan Li1 Xu Sun2 1Beijing Institute of Technology 2MOE Key Laboratory of Computational Linguistics, School of EECS, Peking University 3JD.com {shaoxiongfeng, binsun, likan}@bit.edu.cn {renxc, xusun}@pku.edu.cn ac@chenhongshen.com Abstract Human dialogues are scenario-based and ap- propriate responses generally relate to the la- tent context knowledge entailed by the specific scenario. To enable responses that are more meaningful and context-specific, we propose to improve generative dialogue systems from the scenario perspective, where both dialogue history and future conversation are taken into account to implicitly reconstruct the scenario knowledge. More importantly, the conversa- tion scenarios are further internalized using im- itation learning framework, where the conven- tional dialogue model that has no access to fu- ture conversations is effectively regularized by transferring the scenario knowledge contained in hierarchical supervising signals from the scenario-based dialogue model, so that the fu- ture conversation is not required in actual infer- ence. Extensive evaluations show that our ap- proach significantly outperforms state-of-the- art baselines on diversity and relevance, and expresses scenario-specific knowledge. 1 Introduction Neural dialogue generation has drawn increasing at- tention due to its vast commercial values and practi- cal demands. Typically, given the dialogue history, neural dialogue models, such as plain Seq2Seq model (Sutskever et al., 2014) and Transformer (Vaswani et al., 2017), learn to predict responses via maximum likelihood estimation (Vinyals and Le, 2015; Shang et al., 2015). Different from other sequence generation tasks, such as machine translation and paraphrase genera- tion, the dialogue generation task can be regarded as a loose-coupling task, which has much freedom in the semantic and the linguistic aspects of the generated responses. However, it is often hard for the existing models to handle such freedom, com- pared to the fact that humans have no problem in I’m glad these cheap and cheerful batteries are on sale. I’m sorry. These batteries are not on sale. But the ad said they will be sold for a week. I’m glad these cheap and cheerful batteries are on sale. Unfortunately, all types of batteries are costing more these days. Could you cut the price a little, please? Scenario 1 Scenario 2 But the ad said they will be sold for a week. Could you cut the price a little, please? Figure 1: Examples of one dialogue history and its dif- ferent responses followed by related future conversa- tions. Different responses imply various conversation scenarios, which can be inferred by different relevant future conversations. giving specific yet varied responses even for open- ended dialogue history (Shen et al., 2018; Csaky et al., 2019). One important reason is that we can extend the given dialogue with many possible sce- narios of enriched, imaginative background infor- mation from our experience and world knowledge, to which existing systems have no access. It is beneficial for the dialogue systems to …

Content truncated to 500 words