Tag
This paper proposes a framework for fallacy classification that uses LLMs to extract patterns from fallacious examples and their explanations, achieving statistically significant improvements over zero-shot baselines and demonstrating cross-dataset generalization.