Seamlessly Integrating Factual Information and Social Content with Persuasive Dialogue

Proceedings of the 2nd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 12th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 399–413 November 20–23, 2022. ©2022 Association for Computational Linguistics 399 Seamlessly Integrating Factual Information and Social Content with Persuasive Dialogue Maximillian Chen1, Weiyan Shi1, Feifan Yan1, Ryan Hou1, Jingwen Zhang2, Saurav Sahay3, Zhou Yu1 1Columbia University 2University of California, Davis 3Intel Labs maxchen@cs.columbia.edu {ws2634, fy2241, rh2920, zy2641}@columbia.edu jwzzhang@ucdavis.edu, saurav.sahay@intel.com Abstract Complex conversation settings such as persua- sion involve communicating changes in attitude or behavior, so users’ perspectives need to be addressed, even when not directly related to the topic. In this work, we contribute a novel modular dialogue system framework that seam- lessly integrates factual information and social content into persuasive dialogue. Our frame- work is generalizable to any dialogue tasks that have mixed social and task contents. We con- ducted a study that compared user evaluations of our framework versus a baseline end-to-end generation model. We found our framework was evaluated more favorably in all dimensions including competence and friendliness, com- pared to the end-to-end model which does not explicitly handle social content or factual ques- tions. 1 Introduction Persuasive dialogue systems are designed for chat- bots to communicate with and to influence users with specific goals. Such systems are often de- signed to benefit individual users (e.g., promoting healthy behaviors) or society at large (e.g., per- suading people to make donations). Wang et al. (2019) introduced this idea with the PERSUASION- FORGOOD dataset, which contains 1,017 human- human conversations where one participant per- suaded the other to donate to the charitable organi- zation Save the Children1, with 300 conversations having sentence-level dialogue act annotations. The social and communicative dynamics be- hind persuasive conversation contexts are com- plex. A persuasive conversation by definition in- volves one party, the persuader, intending to change the attitude or behavior of the other party, the persuadee (Torning and Oinas-Kukkonen, 2009). Changing persuadees’ attitude has several dimen- sions including establishing mutual trust and credi- bility, strategically presenting persuasive appeals, 1https://www.savethechildren.net/ Human: Yeah, a lot of developing countries don't have systems or infrastructure to protect children from these causes. That's why organizations like Save the Children are so important. Baseline: I am going to donate a bit of my money to Save the Children. RAP: I agree. There is a lack of support for children in developing countries, especially in war zones. It's so hard to imagine what it's like for a child to grow up facing the daily threat of violence. Figure 1: Chatbot running on the baseline BART model and chatbot running on RAP responding to the same user utterance. The baseline model does not appropriately acknowledge the user’s statement, whereas RAP is able to show acknowledgement and respond appropriately. and eliciting emotional reactions from the per- suadee (O’keefe, 2015; Wilson, 2003). More- over, Grice’s Maxims of Conversation define con- versations as a cooperative and collaborative pro- cess (Grice, 1975; Clark, 1996; Merrison …

Content truncated to 500 words