Unsupervised Knowledge Selection for Dialogue Generation

Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 1230–1244 August 1–6, 2021. ©2021 Association for Computational Linguistics 1230 Unsupervised Knowledge Selection for Dialogue Generation Xiuyi Chen, Feilong Chen, Fandong Meng, Peng Li, Jie Zhou Pattern Recognition Center, WeChat AI, Tencent Inc, Beijing, China {hugheren.chan,ivess.chan}@gmail.com {fandongmeng,patrickpli,withtomzhou}@tencent.com Abstract Knowledge selection is an important and chal- lenging task which could provide the appropri- ate knowledge for informative dialogue gen- eration. However, the needed gold knowl- edge label is difficult to collect in reality. In this paper, we study knowledge selection for dialogue generation in the unsupervised sce- nario and propose a novel Distilled Distant Supervision Loss (DDSL) to supervise knowl- edge selection when the gold knowledge la- bel is unknown. Specifically, we first obtain an oracle knowledge label via distant super- vision and then leverage knowledge distilla- tion to alleviate the noisy labeling problem of distant supervision. Furthermore, we pro- pose a pretraining-finetuning strategy to deal with the mismatch knowledge selection prob- lem that models tend to select the mismatched knowledge for dialogue generation in the un- supervised setting and will cause the degen- eration of knowledge-aware decoder. Exper- iments on two knowledge-grounded dialogue datasets show that our approach manages to se- lect knowledge more accurately in the unsuper- vised setting and generates more informative responses, even outperforming many strong su- pervised baselines.1 1 Introduction To avoid general and dull dialogue generation (Li et al., 2016), knowledge-grounded dialogue which equips dialogue systems with external knowledge has become a popular research topic. Thanks to the hand-collected knowledge-grounded dialogue datasets which align each dialogue (even each utter- ance) with a pre-identified document (Zhang et al., 2018; Zhou et al., 2018; Moghe et al., 2018; Zhou et al., 2020), many researchers focus on inject- ing the given knowledge to generate informative responses and achieve promising results (Yavuz 1The codes and model checkpoints will be available at https://github.com/ErenChan/UKSDG et al., 2019; Tang and Hu, 2019; Qin et al., 2019a; Li et al., 2019b; Zheng and Zhou, 2019; Meng et al., 2019; Ren et al., 2019; Ye et al., 2020; Lin et al., 2020). However, they usually need the pre- identified knowledge and the knowledge access task is less studied (Kim et al., 2020b) which is the precursor to knowledge dialogue generation in reality (Zheng et al., 2020). It is natural to extract the external knowledge via information retrieval technology. Several works re- gard the retrieved knowledge sentences as the pre- identified document (Ghazvininejad et al., 2018; Michel Galley, 2018; Yang et al., 2019b; Gopalakr- ishnan et al., 2019). However, the retrieved docu- ment contains redundant and irrelevant informa- tion which are harmful for dialogue generation (Zhao et al., 2020b). Hence, knowledge selection which chooses an appropriate sentence from the pre-retrieved knowledge pool gains much attention and it plays the role of knowledge access for gener- ation. Dinan et al. (2019) first propose knowledge selection for dialogue generation which are two sequential subtasks and the generation is based on the selected knowledge. Several works follow their …

Content truncated to 500 words