Belief-based generation of Argumentative Claims

When engaging in argumentative discourse, skilled human debaters tailor\r\nclaims to the beliefs of the audience, to construct effective arguments.\r\nRecently, the field of computational argumentation witnessed extensive effort\r\nto address the automatic generation of arguments. However, existing approaches\r\ndo not perform any audience-specific adaptation. In this work, we aim to bridge\r\nthis gap by studying the task of belief-based claim generation: Given a\r\ncontroversial topic and a set of beliefs, generate an argumentative claim\r\ntailored to the beliefs. To tackle this task, we model the people\u0027s prior\r\nbeliefs through their stances on controversial topics and extend\r\nstate-of-the-art text generation models to generate claims conditioned on the\r\nbeliefs. Our automatic evaluation confirms the ability of our approach to adapt\r\nclaims to a set of given beliefs. In a manual study, we additionally evaluate\r\nthe generated claims in terms of informativeness and their likelihood to be\r\nuttered by someone with a respective belief. Our results reveal the limitations\r\nof modeling users\u0027 beliefs based on their stances, but demonstrate the\r\npotential of encoding beliefs into argumentative texts, laying the ground for\r\nfuture exploration of audience reach.