Repairing Fallacious Argumentation in Political Debates
Résumé
Fallacies, defined as arguments based on erroneous logical foundations, are pervasive in political discourse due to their persuasive nature and illusion of validity. Such misleading content and its spread may seriously influence and impact societal well-being by leading to inaccurate conclusions and invalid inferences from citizens and policymakers. Automatically detect and defuse fallacious arguments is therefore crucial to limit the spread of manipulative claims and to promote a healthier political debate. In this paper, we address such challenge, casting it as the computational task of repairing fallacious arguments in political debates. Our contribution in this novel task is manifold. Firstly, we introduce a new dataset, FallacyFix: A Repaired Fallacies Dataset, comprising repaired examples across various fallacy categories. Secondly, we propose a series of prompt techniques for generating non-fallacious arguments, (in)dependent of the fallacy label being addressed. Third, we introduce a novel evaluation methodology to assess the quality of the generated text, specifically designed to repair fallacies in political debates. Lastly, we perform a user study to assess the Relevance, Suitability, and Cogency of the generated repaired arguments.
Domaines
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |