Dogwhistles and Change in Political Meaning: Democratic Discourse in Sweden

• In Event: Misinformation and Marginalization

Abstract

The term dogwhistling denotes conveying often controversial hidden messages through words or expressions, understood by a select group, while going unnoticed by others. Dogwhistles have implications for current discussions on the retrenchment of democracy. The question of whether political actors can use dogwhistles to convey hidden messages to subsets of voters carries important democratic implications. Political dogwhistling has been claimed to erode the legitimacy of a democratic mandate, as it obscures voters’ understanding of their voting options (Goodin & Saward, 2005).

Dogwhistles emerge over time. Until recently, little work has been done on the mechanics of dogwhistle emergence. Insofar as dogwhistles represent a form of strategic ambiguity, it requires that a non-conventional meaning recognized by an “in-group” develops alongside a widely-accepted meaning recognized by an “out-group”. Operationalizing the emergence of meaning has been a domain of research in linguistics, particularly semantics, but advanced techniques from this area are rarely applied to political discourse and media. In our work, we combine very recent approaches from computational semantics, in particular word embeddings and large language models, with innovative techniques in political communication research, which we apply to the Swedish case.

Sweden is a multi-party democracy with proportional representation that has been divided on the matter of immigration and refugee policy resulting in the growth of political trends that question the traditional democratic consensus. The close-knit nature of a media for a country of approx. 10 million residents allows the study of the development of strategic meaning manipulation in popular online media.

Via the Swedish Citizens Panel, our previous study identified several immigration-related dogwhistles, such as återvandring (remigration) and berika (to enrich), which have a widely-understood benign interpretation (voluntary return to one's home country, cultural improvement) and an interpretation in anti-immigration political forums (forced expulsion, migrant crime). In another study, we showed that these have implications for voting and social media endorsement patterns. Here we examine Swedish-language online discussion forums with different orientations towards immigration: Familjeliv (Family Life) and Flashback, forums whose main content is general discussion about living in Swedish society, with the latter having a strong anti-immigrant subtext.

Specifically, our research questions (RQs) are:
* To what extent do the “in-group” meanings of dogwhistles diverge from their “conventional” meanings over time?
* Comparing dogwhistle terms, how different or similar are their patterns of change?
* What are the differences between communities in terms of dogwhistle meaning change?

In linguistics, the Distributional Hypothesis (Harris, 1954) suggests that the meaning of words can be represented by calculating statistics from the words’ surrounding context. This has been implemented in present-day techniques that allow words and phrases to be represented as high-dimensional vectors derived from contextual distributions found in large quantities of text. These vectors can be compared to identify differences in meaning. We showed in Hertzberg et al. (2022) that we can use these vectors to distinguish between “in-group” and “out-group” interpretations of Swedish dogwhistle terms. In parallel, Noble et al. (2021) developed a way to examine semantic change in online communities over time by using corrective factors to align vector representations derived from different training samples.

In this study, we answer our RQs by combining these techniques to quantify and visualize patterns of semantic change among dogwhistles. We find:
* the dogwhistles we study have significantly different patterns of in-group meaning change relative to out-group meaning change.
* patterns of dogwhistle change reflect the media history of those dogwhistles.
* “in-group” meanings of racial dogwhistles are more prominent in the social media forum with an anti-immigrant subtext (Flashback) than in Familjeliv.

We thus show that we can track the development of meanings that challenge an existing democratic consensus using advanced natural language processing techniques.

Goodin, R., & Saward, M. (2005). Dogwhistles and democratic mandates. The Political Quarterly, 76(4), 471–476.

Harris, Z. S. (1954). Distributional structure. Word, 10, 146–162.

Hertzberg, N., Sayeed, A., Breitholtz, E., Cooper, R., Lindgren, E., Rönnerstrand, B., & Rettenegger, G. (2022). Distributional properties of political dogwhistle representations in Swedish BERT. Proceedings of the Sixth Workshop on Online Abuse and Harms (WOAH), 170–175.

Noble, B., Sayeed, A., Fernández, R., & Larsson, S. (2021). Semantic shift in social networks. Proceedings of SEM 2021: The Tenth Joint Conference on Lexical and Computational Semantics, 26–37.

Authors

• Max Boholm, University of Gothenburg

• Asad Sayeed, University of Gothenburg

• Robin Cooper, University of Gothenburg

• Elina Lindgren, University of Gothenburg

• Ellen Breitholtz, University of Gothenburg

• Björn Rönnerstrand