Staying on Topic: An Indicator of Power in Political Debates

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1481–1486, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Staying on Topic: An Indicator of Power in Political Debates Vinodkumar Prabhakaran Dept. of Computer Science Columbia University New York, NY, USA vinod@cs.columbia.edu Ashima Arora Dept. of Computer Science Columbia University New York, NY, USA aa3470@columbia.edu Owen Rambow CCLS Columbia University New York, NY, USA rambow@ccls.columbia.edu Abstract We study the topic dynamics of interac- tions in political debates using the 2012 Republican presidential primary debates as data. We show that the tendency of candidates to shift topics changes over the course of the election campaign, and that it is correlated with their relative power. We also show that our topic shift features help predict candidates’ relative rankings. 1 Introduction The field of computational social sciences has cre- ated many interesting applications for natural lan- guage processing in recent years. One of the areas where NLP techniques have shown great promise is in the analysis of political speech. For example, researchers have applied NLP techniques to polit- ical texts for a variety of tasks such as predicting voting patterns (Thomas et al., 2006), identifying markers of persuasion (Guerini et al., 2008), cap- turing cues that signal charisma (Rosenberg and Hirschberg, 2009), and detecting ideological po- sitions (Sim et al., 2013). Our work also analyzes political speech, more specifically, presidential de- bates. The contribution of this paper is to show that the topic shifting tendency of a presidential candidate changes over the course of the election campaign, and that it is correlated with his or her relative power. We also show that this insight can help computational systems that predict the candi- dates’ relative rankings based on their interactions in the debates. 2 Motivation The motivation for this paper stems from prior work done by the first author in collaboration with other researchers (Prabhakaran et al., 2013a; Prabhakaran et al., 2013b). Prabhakaran et al. (2013a) introduced the notion of power in the do- main of presidential debates, and Prabhakaran et al. (2013b) followed it up with an automatic power ranker system based on interactions within the de- bates. The power that a candidate had at a cer- tain point in the election campaign was modeled based on his or her recent poll standings: in elec- tions, popularity is power. Those studies analyzed the 2012 Republican presidential primary debates and found that a candidate’s power at the time of a debate correlates with the structure of interac- tions within the debate (e.g., turn frequency and interruption patterns). Another finding was that the candidates’ power correlates with the distribu- tion of topics they speak about in the debates: can- didates with more power spoke significantly more about certain topics (e.g., economy) and less about certain other topics (e.g., energy). However, these findings relate to the specific election cycle that was analyzed and will not carry over to political debates in general. A further dimension with relevance beyond a specific …

Content truncated to 500 words