Learning as Conversation: Dialogue Systems Reinforced for Information Acquisition

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4781 - 4796 July 10-15, 2022 ©2022 Association for Computational Linguistics Learning as Conversation: Dialogue Systems Reinforced for Information Acquisition Pengshan Cai,1∗Hui Wan,2 Fei Liu,3 Mo Yu,2 Hong Yu,1,4 Sachindra Joshi2 1Manning College of Information & Computer Sciences, University of Massachusetts, Amherst 2IBM Research AI 3Department of Computer Science, University of Central Florida 4CHORDS, University of Massachuetts, Lowell {pengshancai, hongyu}@cs.umass.edu, {hwan, yum}@us.ibm.com, feiliu@cs.ucf.edu, jsachind@in.ibm.com Abstract We propose novel AI-empowered chat bots for learning as conversation where a user does not read a passage but gains informa- tion and knowledge through conversation with a teacher bot. Our information-acquisition- oriented dialogue system employs a novel adaptation of reinforced self-play so that the system can be transferred to various domains without in-domain dialogue data, and can carry out conversations both informative and atten- tive to users. Our extensive subjective and ob- jective evaluations on three large public data corpora demonstrate the effectiveness of our system to deliver knowledge-intensive and at- tentive conversations and help end users sub- stantially gain knowledge without reading pas- sages. Our code and datasets are publicly avail- able 1 for follow-up research. 1 Introduction Communication is the central process of educa- tion (Dewey, 1923). In learning as conversation (Sharples, 2005), a student does not read a pas- sage but gains information and knowledge through conversation with a teacher who reads the passage. Compared to the traditional learning by reading, learning as conversation has the advantages that conversation helps students stay engaged and that information is provided piece by piece, which helps strengthen learning with a shorter attention span. The advantages of learning as conversation have been verified with educational evidence (Mol et al., 2008; Lever and Sénéchal, 2011; Golinkoff et al., 2019). For example, studies have shown that when children read storybooks, parents’ guided conver- sation, e.g., posing questions and providing respon- sive feedback, substantially amplifies the learning benefits. While high-quality conversations with experts are not always available, it would be help- ∗* Work done during internship at IBM Research AI 1Code and data at: https://github.com/IBM/ reinforced-dialog-system-for-learning ful if AI-empowered chat bots could be applied to facilitate users to gain information or knowledge. Figure 1: The teacher bot educates an user about "BERT" through conversation generated from the document. In recent years, there has been significant re- search in content-grounded dialogue generation, where external passages are used to inspire knowl- edge intensive dialogues. However, these systems or datasets are either for chit chat (Zhou et al., 2018; Dinan et al., 2019) or for goal-oriented informa- tion seeking (Feng et al., 2020; Chen et al., 2021), little work has explored applying chat bots for the learning as conversation purpose. In this work we propose a novel task for learning as conversation: information-acquisition-oriented dialogue generation. Given a passage, our chat bot actively engages with an end-user to form a coherent conversation, so that the user could gain knowledge without reading the …

Content truncated to 500 words