Fallacies of appeal to emotions Corpus: an approach to automatic fallacies identification

Abstract

 Political speeches in electoral campaigns are aimed at mobilizing and attracting the electorate with persuasive messages, and are mainly argued appealing to emotions, committing fallacies. This article presents a fallacies corpus of political speeches made by presidency candidates of Mexico, with the aim obtaining a linguistic resource in Spanish that allows computer development systems for its mining. Until now, there is no known fallacies corpus for Spanish language and arguments corpus elaborated in Argument Mining area are limited to argumentative structure tagging and are not elaborated from political speeches. The corpus was elaborated with arguments extracted from the speeches and a manual annotation of premises and conclusions was made. Inter-annotator agreement of 0.692 was obtained using Cohen's kappa index. Subsequently, valid arguments and fallacies were identified, and 0.442 agreement was obtained with the same index as a result. As an additional contribution, a fallacies identification baseline is presented using cosine similarity, support vector machine, logistic regression and decision trees methods, and effective extraction terms in the arguments.In this baseline, an 0.62 F1-score was obtained and it is a comparison result for future research.