The Moral Debater: A Study on the Computational Generation of Morally Framed Arguments

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 8782 - 8797 May 22-27, 2022 c⃝2022 Association for Computational Linguistics The Moral Debater: A Study on the Computational Generation of Morally Framed Arguments Milad Alshomary † Roxanne El Baff ‡ Timon Gurcke † Henning Wachsmuth † †Paderborn University, Paderborn, Germany, milad.alshomary@upb.de ‡German Aerospace Center, Oberpfaffenhofen, Germany, Roxanne.ElBaff@dlr.de Abstract An audience’s prior beliefs and morals are strong indicators of how likely they will be affected by a given argument. Utilizing such knowledge can help focus on shared values to bring disagreeing parties towards agreement. In argumentation technology, however, this is barely exploited so far. This paper studies the feasibility of automatically generating morally framed arguments as well as their effect on different audiences. Following the moral foun- dation theory, we propose a system that effec- tively generates arguments focusing on differ- ent morals. In an in-depth user study, we ask liberals and conservatives to evaluate the im- pact of these arguments. Our results suggest that, particularly when prior beliefs are chal- lenged, an audience becomes more affected by morally framed arguments. 1 Introduction In the last years, more research has been dedicated to studying prior beliefs in argumentation. Un- derstanding the role of prior beliefs helps people craft more effective arguments when targeting a particular audience. Accordingly, operationalizing knowledge about the audience as part of support- ing tools for humans or realized in fully automated debating technologies (Slonim et al., 2021) could benefit the production of arguments that bridge the gap between disputed parties by focusing on the shared beliefs rather than divisive ones (Alshomary and Wachsmuth, 2021). In social psychology, a body of research em- ploys the notion of morals to understand people’s judgments on controversial topics (Haidt, 2012; Fulgoni et al., 2016). Feinberg and Willer (2015) demonstrated that arguments become more effec- tive when they match the morals of the target audi- ence. Multiple works in computational linguistics have analyzed the persuasive effectiveness of argu- ments depending on the target audience (Durmus and Cardie, 2018; El Baff et al., 2020), showing that audience-based features reliably predict effec- tiveness. Different proxies of beliefs have been proposed as part of this, ranging from interests and personality traits (Al Khatib et al., 2020) to stances on popular issues (Alshomary et al., 2021a). The authors of the latter used the stances to control the generation of argumentative texts. However, they did not assess the effectiveness of these texts on the audience, leaving the importance of encod- ing beliefs ultimately unclear. Beyond that, little research has been done on generating arguments tailored towards a specific audience, let alone on the importance of morals in achieving agreement. This work studies the feasibility of generating morally framed arguments computationally and the effect of these arguments on different audiences. For this, we rely on the moral foundation theory (Haidt, 2012) that projects the moral system into five foundations: care, fairness, loyalty, authority, and purity. To produce arguments that address …

Content truncated to 500 words