Detection and classification of prototypical and spontaneous fallacies in Spanish

The use of fallacies in public debates in political, health, economic and social contexts is detrimental in that it hinders understanding between the parties and facilitates the manipulation of public opinion and the propagation of misinformation. Recently, datasets containing various types of fallacies have become available, allowing experimentation in tasks such as automatic fallacy classification. In this paper, we present the first corpus of fallacies in Spanish, with two distinct sections: one formed by prototypical examples extracted from educational materials, and the other by spontaneous examples extracted from on-line comments to news items. Both sections include examples of non-fallacious texts of similar subject matter. Preliminary results on the detection and classification tasks using the corpus we have created show that it is a challenging task (especially when focused on spontaneous fallacies) that could be a good candidate to be part of the tasks with which the latest advances in language models are evaluated.