And That’s A Fact: Distinguishing Factual and Emotional Argumentation in Online Dialogue

Proceedings of the 2nd Workshop on Argumentation Mining, pages 116–126, Denver, Colorado, June 4, 2015. c⃝2015 Association for Computational Linguistics And That’s A Fact: Distinguishing Factual and Emotional Argumentation in Online Dialogue Shereen Oraby∗, Lena Reed∗, Ryan Compton∗, Ellen Riloff †, Marilyn Walker∗ and Steve Whittaker∗ ∗ University of California Santa Cruz {soraby,lireed,rcompton,mawalker,swhittak}@ucsc.edu † University of Utah riloff@cs.utah.edu Abstract We investigate the characteristics of factual and emotional argumentation styles observed in online debates. Using an annotated set of FACTUAL and FEELING debate forum posts, we extract patterns that are highly correlated with factual and emotional arguments, and then apply a bootstrapping methodology to find new patterns in a larger pool of unanno- tated forum posts. This process automatically produces a large set of patterns representing linguistic expressions that are highly corre- lated with factual and emotional language. Fi- nally, we analyze the most discriminating pat- terns to better understand the defining charac- teristics of factual and emotional arguments. 1 Introduction Human lives are being lived online in transformative ways: people can now ask questions, solve prob- lems, share opinions, or discuss current events with anyone they want, at any time, in any location, on any topic. The purposes of these exchanges are var- ied, but a significant fraction of them are argumenta- tive, ranging from hot-button political controversies (e.g., national health care) to religious interpretation (e.g., Biblical exegesis). And while the study of the structure of arguments has a long lineage in psychol- ogy (Cialdini, 2000) and rhetoric (Hunter, 1987), large shared corpora of natural informal argumenta- tive dialogues have only recently become available. Natural informal dialogues exhibit a much broader range of argumentative styles than found in traditional work on argumentation (Marwell and Schmitt, 1967; Cialdini, 2000; McAlister et al., 2014; Reed and Rowe, 2004). Recent work has be- gun to model different aspects of these natural in- formal arguments, with tasks including stance clas- sification (Somasundaran and Wiebe, 2010; Walker et al., 2012), argument summarization (Misra et al., 2015), sarcasm detection (Justo et al., 2014), and work on the detailed structure of arguments (Bi- ran and Rambow, 2011; Purpura et al., 2008; Yang and Cardie, 2013). Successful models of these tasks have many possible applications in sentiment detection, automatic summarization, argumentative agents (Zuckerman et al., 2015), and in systems that support human argumentative behavior (Rosenfeld and Kraus, 2015). Our research examines FACTUAL versus FEELING argument styles, drawing on annotations provided in the Internet Argument Corpus (IAC) (Walker et al., 2012). This corpus includes quote-response pairs that were manually annotated with respect to whether the response is primarily a FACTUAL or FEELING based argument, as Section 2.1 describes in more detail. Figure 1 provides examples of re- sponses in the IAC (paired with preceding quotes to provide context), along with the response’s FAC- TUAL vs. FEELING label. FACTUAL responses may try to bolster their ar- gument by providing statistics related to a position, giving historical or scientific background, or pre- senting specific examples or data. There is clearly a …

Content truncated to 500 words