Yes, we can! Mining Arguments in 50 Years of US Presidential Campaign Debates

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4684–4690 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 4684 Yes, we can! Mining Arguments in 50 Years of US Presidential Campaign Debates Shohreh Haddadan University of Luxembourg Luxembourg shohreh.haddadan@uni.lu Elena Cabrio and Serena Villata Universit´e Cˆote d’Azur, CNRS, Inria, I3S France firstname.surname@unice.fr Abstract Political debates offer a rare opportunity for citizens to compare the candidates’ positions on the most controversial topics of the cam- paign. Thus they represent a natural applica- tion scenario for Argument Mining. As ex- isting research lacks solid empirical investiga- tion of the typology of argument components in political debates, we fill this gap by propos- ing an Argument Mining approach to politi- cal debates. We address this task in an em- pirical manner by annotating 39 political de- bates from the last 50 years of US presidential campaigns, creating a new corpus of 29k ar- gument components, labeled as premises and claims. We then propose two tasks: (1) iden- tifying the argumentative components in such debates, and (2) classifying them as premises and claims. We show that feature-rich SVM learners and Neural Network architectures out- perform standard baselines in Argument Min- ing over such complex data. We release the new corpus USElecDeb60To16 and the ac- companying software under free licenses to the research community. 1 Introduction Political debates are public interviews where the candidates of elections are requested to confront each other on topics such as unemployment, taxes, and foreign policy. During presidential elections in the US, it is customary for the main candidates of the two largest parties, i.e., the Democratic and the Republican Parties, to engage in a debate around the most controversial issues of the time. Such debates are considered as a de facto elec- tion process, and in some cases they have nearly decided the outcomes of the elections (Coleman et al., 2015). Given the importance of these debates and their innate argumentative features, they represent a natural playground for Argument(ation) Min- ing (AM) methods (Peldszus and Stede, 2013; Lippi and Torroni, 2016b; Cabrio and Villata, 2018). AM deals with analyzing argumentation in various domains, such as legal cases (Mochales and Moens, 2011), persuasive essays (Stab and Gurevych, 2017), clinical trials (Mayer et al., 2018) and scientific articles (Teufel et al., 2009). The ability of identifying argumentative compo- nents and predicting their relations in such a kind of texts opens the door to cutting-edge tasks like fallacy detection, fact-checking, and counter- argumentation generation. Despite the plethora of existing approaches and annotated corpora for AM, very few of them tackle the issue of mining argumentative structures from political debates (Lippi and Torroni, 2016a; Menini et al., 2018; Duthie and Budzynska, 2018; Visser et al., 2019). To be best of our knowledge, none of them take on the identification of argu- ment components (i.e., premises and claims) on a large corpus of political debates. This paper fills this gap by (1) performing a large-scale …

Content truncated to 500 words