Leveraging Linguistic Content and Debater Traits to Predict Debate Outcomes.

Leveraging Linguistic Content and Debater Traits to Predict Debate Outcomes Alexandra Paxton (paxton.alexandra@gmail.com) Rick Dale (rdale@ucmerced.edu) Cognitive & Information Sciences, University of California, Merced 5200 N. Lake Road, Merced, CA 95343 Abstract Since the earliest televised debates, cognitive and political sciences have been interested in how voters respond to political candidates and their messages, both verbal and nonverbal. The present work draws from this long tradition and combines it with work on persuasion and rhetoric to inform analyses of a new corpus of debate data: 48 transcripts from the Intelligence Squared U.S. series, televised Oxford-style debates on relevant sociopolitical issues (http://www.iq2us.org). As a first look at this corpus, we focus on how linguistic content (i.e., hedging and pronoun use) and debater traits (i.e., attractiveness and negativity) interact with arbitrary group identity (i.e., “for” vs. “against”) to affect debate outcomes. Interestingly, we find that arbitrary group identity (i.e., “for” vs. “against” labels created by the framing of the debate rather than the actual opinions held) significantly affects the ways in which linguistic content and debater traits influence voters. Keywords: communication; conflict; corpus analysis; debate; persuasion; politics; political psychology; political science Introduction Conflict is a regular part of the human experience. From legal battles to quarrels over chores, we regularly deal with conflict on personal, national, and international scales. While we may not necessarily enjoy these conflicts, we generally recognize that they are an essential part of our social experience. In fact, on a cultural level, it could be argued that we very highly value conflict in its proper place. Many democratic nations have adversarial judicial systems, requiring parties involved in legal action to argue their cases at the expense of the other, and hold debates as a key element of the electoral process. In light of the importance of conflict, it is hardly surprising that so many have undertaken to try to explain it. Philosophers, political scientists, and cognitive scientists have attempted to answer questions of the origins of conflict, its purpose, and its essential characteristics using a variety of methods. The current project attempts to unite these perspectives to drive investigations of naturalistic debate using a newly compiled corpus of debate transcripts among experts on socially relevant topics. We are specifically interested in investigating the ways in which debaters affect one another and their audience. With its interdisciplinary approach and computational focus, cognitive science is poised to uniquely and substantively add to our understanding of the topic. In the present work, we hope to spark such investigations by blending ideas from political science and pragmatics to shape linguistic analysis of a novel corpus of Oxford-style sociopolitical debates. Persuasion, Political Science, and Pragmatics Political science has been particularly enamored with debates since their first televised appearance (e.g., Baker & Norpoth, 1981). Related political and psychological research has investigated the effects of political advertising and campaigning (e.g., Geer, 2008). Both of these lines of research tend to incorporate an interest in the individual differences within the audience and in the manner in which the message is delivered. Given the nature of the corpus, we will focus more on the latter, although our hypotheses and analyses will also be shaped by ideas about the audience. Voter Characteristics and Tendencies Integrating decades of reasoning studies, cognitive scientists Mercier and Sperber (2011) have recently suggested that human cognition appears to be geared toward argumentation and the defense of personal beliefs rather than reason. This view is consistent with a thread of findings in political science. While much of this work uncovers individual differences in the characteristics that drive political behavior, opinion, and opinion change (e.g., Brandt et al., 2014; Jost, Glaser, Kruglanski, & Sulloway, 2003; Koch, 1998), a recurring finding suggests that voters engage in motivated political reasoning and have a strong confirmation bias. Highly politically informed individuals tend to be swayed less than moderately or poorly informed individuals (Koch, 1998), and mass media and political advertising – while influential among undecided voters – are substantially less likely to change voters’ opinions, once made (Forrest & Marks,