Adding Chit-Chat to Enhance Task-Oriented Dialogues

Existing dialogue corpora and models are typically designed under two disjoint motives: while task-oriented systems focus on achieving functional goals (e.g., booking hotels), open-domain chatbots aim at making socially engaging conversations. In this work, we propose to integrate both types of systems by Adding Chit-Chat to ENhance Task-ORiented dialogues (ACCENTOR), with the goal of making virtual assistant conversations more engaging and interactive. Specifically, we propose a Human ↔AI collaborative data collection approach for generating diverse chitchat responses to augment task-oriented dialogues with minimal annotation effort. We then present our new chit-chat-based annotations to 23.8K dialogues from two popular task-oriented datasets (Schema-Guided Dialogue and MultiWOZ 2.1) and demonstrate their advantage over the originals via human evaluation. Lastly, we propose three new models for adding chit-chat to task-oriented dialogues, explicitly trained to predict user goals and to generate contextually relevant chit-chat responses.

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 systems focuses on a particular type of dialogue system. Work on task-oriented dialogue systems typically aims to track user goals with higher accuracy 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 usually the target of research on chit-chat systems (Li et al., 2019).

Discussion / Conclusion. We propose adding chit-chat to enhance taskoriented dialogues (ACCENTOR) in this study. We present a general Human↔AI collaborative data construction approach for ACCENTOR, with which we create a dataset consisting of 23.8K chit-chat augmented task-oriented dialogues. We show via human evaluation that chit-chat augmented dialogues are preferred than the unaugmented. In addition, we propose three models for ACCENTOR. Evaluation results show that compared with the baseline trained on the original unaugmented data, our proposed models trained on the chit-chat augmented counterpart achieve a similar task performance level and higher human evaluation scores.