Before Name-Calling: Dynamics and Triggers of Ad Hominem Fallacies in Web Argumentation

Proceedings of NAACL-HLT 2018, pages 386–396 New Orleans, Louisiana, June 1 - 6, 2018. c⃝2018 Association for Computational Linguistics Before Name-calling: Dynamics and Triggers of Ad Hominem Fallacies in Web Argumentation Ivan Habernal† Henning Wachsmuth‡ Iryna Gurevych† Benno Stein‡ † Ubiquitous Knowledge Processing Lab (UKP) and Research Training Group AIPHES Department of Computer Science, Technische Universit¨at Darmstadt, Germany www.ukp.tu-darmstadt.de www.aiphes.tu-darmstadt.de ‡ Faculty of Media, Bauhaus-Universit¨at Weimar, Germany <firstname>.<lastname>@uni-weimar.de Abstract Arguing without committing a fallacy is one of the main requirements of an ideal debate. But even when debating rules are strictly enforced and fallacious arguments punished, arguers of- ten lapse into attacking the opponent by an ad hominem argument. As existing research lacks solid empirical investigation of the ty- pology of ad hominem arguments as well as their potential causes, this paper fills this gap by (1) performing several large-scale annota- tion studies, (2) experimenting with various neural architectures and validating our work- ing hypotheses, such as controversy or reason- ableness, and (3) providing linguistic insights into triggers of ad hominem using explainable neural network architectures. 1 Introduction Human reasoning is lazy and biased but it per- fectly serves its purpose in the argumentative con- text (Mercier and Sperber, 2017). When chal- lenged by genuine back-and-forth argumentation, humans do better in both generating and evaluat- ing arguments (Mercier and Sperber, 2011). The dialogical perspective on argumentation has been reflected in argumentation theory prominently by the pragma-dialectic model of argumentation (van Eemeren and Grootendorst, 1992). Not only sketches this theory an ideal normative model of argumentation but also distinguishes the wrong ar- gumentative moves, fallacies (van Eemeren and Grootendorst, 1987). Among the plethora of prototypical fallacies, notwithstanding the contro- versy of most taxonomies (Boudry et al., 2015), ad hominem argument is perhaps the most famous one. Arguing against the person is considered faulty, yet is prevalent in online and offline dis- course.1 1According to ‘Godwin’s law’ known from the in- ternet pop-culture (https://en.wikipedia.org/wiki/ Although the ad hominem fallacy has been known since Aristotle, surprisingly there are very few empirical works investigating its properties. While Sahlane (2012) analyzed ad hominem and other fallacies in several hundred newspaper edi- torials, others usually only rely on few examples, as observed by de Wijze (2002). As Macagno (2013) concludes, ad hominem arguments should be considered as multifaceted and complex strate- gies, involving not a simple argument, but sev- eral combined tactics. However, such research, to the best of our knowledge, does not exist. Very little is known not only about the feasibility of ad hominem theories in practical applications (the NLP perspective) but also about the dynamics and triggers of ad hominem (the theoretical counter- part). This paper investigates the research gap at three levels of increasing discourse complexity: ad hominem in isolation, direct ad hominem with- out dialogical exchange, and ad hominem in large inter-personal discourse context. We asked the following research questions. First, what qualita- tive and quantative properties do ad hominem ar- guments have in Web debates and how does that reflect …

Content truncated to 500 words