SalesBot: Transitioning from Chit-Chat to Task-Oriented Dialogues

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 6143 - 6158 May 22-27, 2022 c⃝2022 Association for Computational Linguistics SalesBot: Transitioning from Chit-Chat to Task-Oriented Dialogues Ssu Chiu† Maolin Li⋆ Yen-Ting Lin† Yun-Nung Chen† †National Taiwan University, Taipei, Taiwan ⋆MediaTek Research, Cambridge, UK r09944026@csie.ntu.edu.tw mmaolin.li@mtkresearch.com f08944064@csie.ntu.edu.tw y.v.chen@ieee.org Abstract Dialogue systems are usually categorized into two types, open-domain and task-oriented. The first one focuses on chatting with users and making them engage in the conversations, where selecting a proper topic to fit the dia- logue context is essential for a successful dia- logue. The other one focuses on a specific task instead of casual talks, e.g., finding a movie on Friday night, playing a song. These two direc- tions have been studied separately due to their different purposes. However, how to smoothly transition from social chatting to task-oriented dialogues is important for triggering the busi- ness opportunities, and there is no any pub- lic data focusing on such scenarios. Hence, this paper focuses on investigating the conver- sations starting from open-domain social chat- ting and then gradually transitioning to task- oriented purposes, and releases a large-scale dataset with detailed annotations for encour- aging this research direction. To achieve this goal, this paper proposes a framework to auto- matically generate many dialogues without hu- man involvement, in which any powerful open- domain dialogue generation model can be eas- ily leveraged. The human evaluation shows that our generated dialogue data has a natural flow at a reasonable quality, showing that our released data has a great potential of guiding future research directions and commercial ac- tivities. Furthermore, the released models al- low researchers to automatically generate un- limited dialogues in the target scenarios, which can greatly benefit semi-supervised and unsu- pervised approaches.1 1 Introduction Until now, researchers have often separated open- domain and task-oriented dialogues as two distinct types of tasks in the dialogue field. The publicly available datasets focuses on either open-domain 1Our dataset, trained simulators, and annotations are avail- able at: https://github.com/MiuLab/SalesBot. Simulated User Simulated Salesperson My son is 2 and a half. He is handful, but I love him so much. No, I don’t. I work too much to have time for one. How about you? I work at a library. It’s a lot of fun, and I get to meet lots of interesting people. I like to read a lot. I also like to go to the movies. What about yourself? That’s adorable. I have a 2-year-old son and he’s the same way. Do you have any pets? I work a lot too. I’m a stay at home mom. What do you do for work? Are you interested in watching any movie? That sounds like a fun job. I bet you meet some interesting people at the library. What else do you like to do? I’m looking for a move to watch. A regular showing would be fine. I’m in San Francisco, please look for movies there. Where are you …

Content truncated to 500 words