Testing and Comparing Computational Approaches for Identifying the Language of Framing in Political News

Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 1472–1482, Denver, Colorado, May 31 – June 5, 2015. c⃝2015 Association for Computational Linguistics Testing and Comparing Computational Approaches for Identifying the Language of Framing in Political News Eric P. S. Baumer1,2, Elisha Elovic2, Ying “Crystal” Qin2, Francesca Polletta3, Geri K. Gay1,2 1Communication 2Information Science 3Sociology Cornell University University of California, Irvine Ithaca, NY, USA Irvine, CA, USA {ericpsb, epe9, yq37, gkg1}@cornell.edu, polletta@uci.edu Abstract The subconscious influence of framing on per- ceptions of political issues is well-document in political science and communication re- search. A related line of work suggests that drawing attention to framing may help reduce such framing effects by enabling frame reflec- tion, critical examination of the framing un- derlying an issue. However, definite guid- ance on how to identify framing does not ex- ist. This paper presents a technique for iden- tifying frame-invoking language. The paper first describes a human subjects pilot study that explores how individuals identify fram- ing and informs the design of our technique. The paper then describes our data collection and annotation approach. Results show that the best performing classifiers achieve perfor- mance comparable to that of human annota- tors, and they indicate which aspects of lan- guage most pertain to framing. Both technical and theoretical implications are discussed. 1 Introduction Contentious political issues are rarely understood per se but rather through the lens of framing. Terms such as “tax relief,” “death taxes,” “racial quotas,” “death panels,” and others have famously rallied cit- izens around fairly complex causes. More gener- ally, research has shown that the way an issue is framed (Entman, 1993) – how a problem is defined, to what other problems and people it is linked, etc. – has a significant impact on both perceptions of the issue and prescriptions for action. A variety of work has shown that minor changes in language – “global warming” vs. “climate change” (Schuldt et al., 2011), “gay civil unions” vs. “homosexual mar- riage” (Price et al., 2005), “not allow” vs. “forbid” (Rugg, 1941) – can significantly impact opinions. A related but less explored line of research suggests that “frame reflection” (Sch¨on and Rein, 1994), i.e., critical thinking about an issue’s fram- ing, can play an important role in understanding is- sues and reconciling conflicts. Indeed, some recent work suggests that drawing attention to framing may help mitigate framing effects (Baumer et al., 2015). However, such reflection is no mean feat. “Various observers have noted how subtly and unconsciously p.[framing] operates” (Gamson and Modigliani, 1989, 7), making it difficult to acknowledge that an issue is being framed at all, let alone examine that framing critically or consider alternatives. Further- more, “straightforward guidelines on how to identify [...] a frame in communication do not exist” (Chong and Druckman, 2007, p. 106). To address this challenge, this paper compares different computational approaches for identifying the language of framing, specifically in political news coverage. The best performing classifiers achieve accuracy around …

Content truncated to 500 words