How Metaphors Impact Political Discourse: A Large-Scale Topic-Agnostic Study Using Neural Metaphor Detection

Metaphors are widely used in political rhetoric as an effective framing device. While the efficacy of specific metaphors such as the war metaphor in political discourse has been documented before, those studies often rely on small number of hand-coded instances of metaphor use. Larger-scale topic-agnostic studies are required to establish the general persuasiveness of metaphors as a device, and to shed light on the broader patterns that guide their persuasiveness. In this paper, we present a large-scale data-driven study of metaphors used in political discourse. We conduct this study on a publicly available dataset of over 85K posts made by 412 US politicians in their Facebook public pages, up until Feb 2017. Our contributions are threefold: we show evidence that metaphor use correlates with ideological leanings in complex ways that depend on concurrent political events such as winning or losing elections; we show that posts with metaphors elicit more engagement from their audience overall even after controlling for various socio-political factors such as gender and political party affiliation; and finally, we demonstrate that metaphoricity is indeed the reason for increased engagement of posts, through a fine-grained linguistic analysis of metaphorical vs. literal usages of 513 words across 70K posts. Metaphorical expressions arise in the presence of systematic metaphorical associations, or conceptual metaphors, mapping one concept or domain to another (Lakoff and Johnson 1980). For instance, when we talk about “curing juvenile delinquency” or “diagnosing corruption”, we view crime (the target domain) in terms of a disease (the source domain) and map various elements of the disease knowledge system to our reasoning about crime. Since recognizing metaphorical expressions is critical in order to correctly interpret their intended meanings, computational approaches to metaphor detection has long been an active area of research in Natural Language Processing (NLP) (Shutova 2010). While recent research has achieved great advances in automatic metaphor detection performance (Gutiérrez et al. 2016; Bulat, Clark, and Shutova 2017; Rei et al. 2017; Gao et al. 2018; Dankers et al. 2019), not much research has investigated whether Copyright © 2021, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. metaphor detection, when employed at scale, could help answer some of the fundamental questions in social sciences about metaphors and their ability to shape the public discourse. Social scientists have demonstrated the importance of metaphors in political rhetoric — their role as a framing technique (Lakoff 1991; Tannen 1993; Entman 2003), their tendency to elicit stronger emotions than their literal counterparts (Citron and Goldberg 2014), and their effectiveness in influencing decision-making (Thibodeau and Boroditsky 2011). For instance, metaphors help with the framing of an issue by selecting and emphasising its facets that reinforce a particular point of view (Lakoff 1991; Musolff 2000; Lakoff and Wehling 2012). Discussing war as a competitive game emphasizes the victory vs. defeat aspect of war, while neglecting its human cost, a strategy politicians could use to arouse a pro-war sentiment in the public (Lakoff 1991). While the role metaphors play in shaping the public discourse has been documented before, these qualitative studies are often limited to a few specific domains or topics (e.g., the war domain above) and the metaphors used in them. In this paper, aided by a neural network based metaphor detection approach, we present a large-scale domainagnostic study of the effects of metaphors in political discourse. Our study is conducted on a dataset of 85K Facebook status message posts made by 412 US politicians over a period of around nine years. Our three main findings are: • We find that politicians’ rate of metaphor use is correlated with their ideological leanings in complex ways that depend on contemporaneous political events. Specifically, our analysis found that Democratic politicians used significantly more metaphors during the three months immediately after the 2016 election loss, compared to themselves in prior months, as well as to Republican politicians during the same period of time or prior months. • We find that politicians’ posts that include metaphorical language tend to have significantly higher engagement from their audience. The scale of our study enabled us to control for various socio-political factors, such as gender and party affiliation, strengthening our findings compared to prior studies. • We also conduct a finer-grained linguistic analysis that reveals (1) that metaphorical uses of source domain words ar X iv :2 10 4. 03 92 8v 1 [ cs .C L ] 8 A pr 2 02 1 lead to a higher engagement than their literal usage, and (2) aspects of target domains (i.e. specific political issues) that receive a greater engagement when described metaphorically, as compared to literally. To our knowledge, this is the first computational study of this scale on the impact of metaphor use in political communication in a topic-agnostic manner.