Adding Chit-Chat to Enhance Task-Oriented Dialogues

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 1570–1583 June 6–11, 2021. ©2021 Association for Computational Linguistics 1570 Adding Chit-Chat to Enhance Task-Oriented Dialogues Kai Sun 1∗, Seungwhan Moon2, Paul Crook2, Stephen Roller3, Becka Silvert2, Bing Liu2, Zhiguang Wang2, Honglei Liu2, Eunjoon Cho2, and Claire Cardie1 1Cornell University 2Facebook, 3Facebook AI Research � ks985@cornell.edu, shanemoon@fb.com Abstract Existing dialogue corpora and models are typ- ically designed under two disjoint motives: while task-oriented systems focus on achiev- ing functional goals (e.g., booking hotels), open-domain chatbots aim at making socially engaging conversations. In this work, we pro- pose to integrate both types of systems by Adding Chit-Chat to ENhance Task-ORiented dialogues (ACCENTOR), with the goal of mak- ing virtual assistant conversations more en- gaging and interactive. Specifically, we pro- pose a Human ↔ AI collaborative data col- lection approach for generating diverse chit- chat responses to augment task-oriented dia- logues with minimal annotation effort. We then present our new chit-chat-based anno- tations to 23.8K dialogues from two popu- lar task-oriented datasets (Schema-Guided Di- alogue and MultiWOZ 2.1) and demonstrate their advantage over the originals via human evaluation. Lastly, we propose three new mod- els for adding chit-chat to task-oriented dia- logues, explicitly trained to predict user goals and to generate contextually relevant chit-chat responses. Automatic and human evaluations show that, compared with the state-of-the-art task-oriented baseline, our models can code- switch between task and chit-chat to be more engaging, interesting, knowledgeable, and hu- manlike, while maintaining competitive task performance. 1 Introduction With modeling innovations, increasing computing power, and a growing number of datasets, recent years have witnessed significant improvements in the performance of both task-oriented dialogue systems and chit-chat systems (Adiwardana et al., 2020; Roller et al., 2020; Hosseini-Asl et al., 2020; Peng et al., 2020a). Most research on dialogue ∗Work done as a research intern at Facebook. The code and data are available at https://github.com/ facebookresearch/accentor. I’m looking for a concert in Vancouver. I found an event for the Beach Boys at PNE Amphitheatre. When does the event start, and what’s the event category? It’s a Pop event starting at 6:30 pm. It’s a great way to kick off the summer! That sounds great! Chit-chat Task-oriented USER SYSTEM Figure 1: A sample task-oriented dialogue snippet aug- mented by chit-chat. systems focuses on a particular type of dialogue system. Work on task-oriented dialogue systems typically aims to track user goals with higher ac- curacy to better achieve functional goals (Rastogi et al., 2020) with the sacrifice of not paying explicit attention to user experience, such as making the conversation more engaging, while the latter is usu- ally the target of research on chit-chat systems (Li et al., 2019). In this work, we step forward and pro- pose to integrate both types of systems by Adding Chit-Chat to ENhance Task-ORiented dialogues (ACCENTOR), aiming to have a virtual assistant ca- pable not only of performing various complex tasks such as checking the weather, …

Content truncated to 500 words