A political dogwhistle is a word or phrase that is used to broadcast a message widely considered socially unacceptable by concealing it inside a term with a conventional meaning. The socially unacceptable meaning is used to attract an "in-group" of voters already aligned with radical or extreme ideologies while remaining palatable to an "out-group" of voters with more moderate views who could be offended by the concealed meaning. For example, "inner city" used to be a dogwhistle in American politics, when it had an "in-group" meaning referring to presumed crime by African-Americans and a more conventional "out-group" meaning referring to the geography of cities. It declined in use as a dogwhistle when too much of the public caught on to the "in-group" meaning.
A lexical replacement task is a means of identifying word senses by asking human participants to replace a word in a sentence context with another phrasing that maintains the participants' understanding of that sentence. We administered a lexical replacement experiment to a sample of the Swedish population through the Swedish Citizens Panel focusing on candidate dogwhistles, and found that the dogwhistles that had the strongest effect of being identified by an in-group member and not an out-group member had to do with the European immigration debate. We then used corpora derived from two active Swedish web forums, Flashback and Familjeliv, the former known for being tolerant to racially discriminatory extremism, to train embedding spaces with the aim of applying lexical semantic change (LSC) analysis to identify the conditions of dogwhistle emergence and expiration. We defined measures of semantic "intensity" and "separation" in the embedding space to enable qualitative analysis and, eventually, retrospective dogwhistle pattern detection in social media. Finally, we returned to the Swedish Citizens Panel with a survey of voter reactions to dogwhistle language and found that dogwhistles succeed at attracting in-group voters without alienating out-group voters.
Asad Sayeed is Associate Professor of Computational Linguistics at the University of Gothenburg, Sweden, and an alumnus of the CLIP lab, receiving his Ph.D. in Computer Science from the University of Maryland, College Park, in 2011. He currently leads the Gothenburg Research Initiative for Politically Emergent Systems (GRIPES), funded by the Marianne and Marcus Wallenberg Foundation as part of the Wallenberg Autonomous Systems Program: Humanity and Society.