The carefully crafted language used by politicians often involves expressions known as dogwhistles. Dogwhistles are expressions that convey two messages: one that is neutral and one that is controversial or norm-violating, and where the latter is designed to appeal to specific subsets of the speaker’s audience while remaining unnoticed by others. In this paper, I argue against Khoo’s (2017) argument that dogwhistles cannot be ambiguous expressions, an argument which many have used in order to motivate fully pragmatic theories of how dogwhistles work. The secondary aim is to develop a tangible ambiguity view of dogwhistles that accommodates recent empirical findings on dogwhistles’ community specific meanings.

A paradigm example of a dogwhistle is “inner city.” During an interview on “Bill Bennet’s Morning in America,” Congressman Paul Ryan addressed poverty and welfare reforms by saying:

We have got this tailspin of culture, in our inner cities in particular, of men not working and just generations of men not even thinking about working or learning the value and the culture of work, and so there is a real culture problem here that has to be dealt with. (Lowery, 2014, bold is added)

In the literature, “inner cities” is ascribed the general meaning “central section of a city,” but is also understood as conveying something about the race and status of the residents of the urban locations that are addressed, namely that one is talking about poor African American city neighborhoods. Khoo (2017) argues that “inner city” cannot be ambiguous because the expression fails tests for ambiguity such as the contradiction test. From this, Khoo contends that dogwhistles in general do not work via ambiguity. However, it is well known that not all ambiguous expressions reliably pass ambiguity tests such as the one Khoo uses. Polysemes, which are ambiguous expressions with two related meanings, are known to fail tests for ambiguity (Viebahn 2018; Geeraerts 1993 and Gillon 2004). I show that dogwhistles with related meanings systematically fail the contradiction test (e.g., “inner city” and “cultural enrichment”) whereas dogwhistles with unrelated meanings actually pass the test (e.g., “jogger” and “skittles”.) What we end up with are highly variable predictions, and the conclusion I draw is that the test underlying Khoo’s argument is unreliable, and that we ought to remain open to the possibility that dogwhistles may work via ambiguity.

A recent empirical study by Boholm & Sayeed (2023) shows that certain Swedish dogwhistles undergo community-specific semantic shifts. By analyzing Swedish online forums, Boholm and Sayeed show that on forums linked with controversial ideologies, users consistently and systematically use certain terms with a dogwhistle meaning whereas users on other forums almost exclusively use the same terms with their neutral meaning. To accommodate these findings, I propose a revamped ambiguity view. Unlike prevailing pragmatic explanations, I contend that many dogwhistles exhibit semantic ambiguity between a widely accepted meaning and a secondary, controversial one. The dogwhistle meaning arises from pertinent sub-communities developing new semantic conventions for word forms already used by other communities. Politicians, for instance, can exploit this phenomenon to garner support from fringe sub-communities without overtly committing to having meant what the dogwhistle means in the vocabulary of the relevant sub-community or to what that community represent (e.g., an ideology or some behaviors that are norm-violating or frowned upon). Theoretical support for this claim will be drawn from Lewis’ (1975) notion of a convention and Clark’s (1996) notion of a communal lexicon, which is a set of word-conventions specific to a sub-community.

Additionally, by adhering to empirical research on ambiguity, I will provide a detailed explanation of when dogwhistles are successful. The empirical research suggest that which reading the interpreter of ambiguous terms will go for is relative to the priming effects present in the context of utterance. I will use this to explain how dogwhistles’ capacity to retain certain discourse effects is relative to the priming in the utterance contexts. It suggests that generally agreed upon features of dogwhistles, such as them being plausible deniable, will most likely only be present in an utterance context where both meanings are compatible with the context of utterance.

REFERENCES
Boholm, M. and Sayeed, A. (2023). Political dogwhistles and community divergence in semantic change. In Proceedings of the 4th Workshop on Computational Approaches to Historical Language Change, pp. 53–65, Singapore. Association for Computational Linguistics.
Clark. H. H. (1996). Using Language. Cambridge University Press.
Gillon B.S. (2004). Ambiguity, indeterminacy, deixis, and vagueness: evidence and theory. In: Davis S, Gillon B.S. (eds) Semantics: a reader. Oxford University Press, Oxford: 157–190
Geeraerts. D. (1993). Vagueness’s puzzles, polysemy’s vagaries. Cogn Linguist 4(3): 223–272
Khoo, J. (2017). Code words in political discourse. Philosophical topics, 45(2): 33–64.
Lewis, D. (1969). Convention: A philosophical study. Harvard University Press.
Lowery, W. (2014). https://www.washingtonpost.com/news/the-fix/wp/2014/03/18/paul-ryanpoverty-dog-whistles-and-racism/ Accessed 3rd of February 2024.
Viebahn, E. (2018) Ambiguity and Zeugma. Pacific Philosophical Quarterly, 99: 749– 762.

People who explicitly endorse progressive and egalitarian values sometimes act in prejudiced ways towards members of stigmatized groups. This is often explained by appeal to implicit biases. The thought is, roughly, that people may explicitly endorse anti-racist and feminist beliefs, and yet have implicit attitudes that contradict these beliefs. This psychological understanding of implicit bias has been criticized by those who maintain that it neglects the structural aspects of the phenomenon. However, a problem with structural explanations is that they are more difficult to grasp; it is unclear what the nature of structures are and whether their existence goes beyond individuals – and if so, how?

In this paper, I offer a structural pragmatic account of the mechanisms behind what I call conversational implicit bias. These include the biases that members of stigmatized groups themselves experience in conversational interactions and the ways that utterances about members of stigmatized groups enforce biases. For instance, consider the case of a mother who returns from parental leave and the first thing her colleague says is “Are you back already?”. We can imagine that this colleague explicitly endorses feminist values, but nevertheless makes this type of utterance which underscores a stereotype that women should prioritize family over career. Another example is Fricker’s influential work on testimonial injustice, where the central idea is that a speakers’ utterance is assigned less credibility because of a prejudiced stereotype about her social identity held by the listener (Fricker, 2007, p. 2). The account of implicit bias in conversation presented here will challenge the last part of this characterization of testimonial injustice, calling into question what it means for the stereotype to be held by the listener.

The account of conversational implicit bias presented builds on dynamic pragmatics, which is broadly the study of the assumptions that interlocutors make about their communicative exchange, such as the meanings of the words they use and the purposes of their conversation. One of the most influential versions of this is Stalnaker’s view that interlocutors make pragmatic presuppositions about what is mutually believed. The set of propositions that are mutually accepted to be common belief form the common ground. This way of modelling a context can be understood as a rational reconstruction of what happens in conversations. When a proposition is included in the common ground, this means that the proposition is henceforth accepted as common belief, where ‘acceptance’ is understood as a sociolinguistic disposition to behave in conversation as if the proposition is mutually believed to be true. Interlocutors can accept a proposition for the sake of conversation – without actually believing it – in which case they will be expected to adjust their behavior in accordance with it (Stalnaker, 1999, pp. 52, 58).

By drawing from Levinson’s (2000, 37) pragmatic view, I argue that interlocutors accept meta-linguistic stereotypes as default presuppositions in the common ground. For instance, because interlocutors mutually accept the stereotype that tomatoes are red, they tend to say things such as “We need tomatoes” not “We need red tomatoes” because it is mutually accepted that the object described instantiates the stereotype, unless one suggests otherwise.

The main hypothesis pursued in this paper is that since stereotypes consist of generic propositions, they inherit two problematic features of generic propositions that are about members of social groups.

First, according to Leslie (2015), we tend to think of social kinds as essentialized which means that we attribute to them a shared underlying nature. Haslanger (2012) argues that generics such as “Women are submissive” or “Boys don’t cry” implicate that there is something about the nature of women that make them submissive and something about the nature of boys that they do not cry. Meta-linguistic stereotypes about social identity groups inherit this tendency to essentialize from the generic propositions that they consist of.

Second, generics about social groups have a normative force (Leslie, 2015). Roughly, the thought is that a generic such as “Boys don’t cry” says something about how boys ought to behave, not only how they do behave. According to Leslie, the normative force associated with generics such as “Boys don’t cry” or “Women are submissive” stem from the social roles that we associate with some social kinds (such as boys and women) and that members of this social kind have a prima facie obligation to possess the features that are constitutive of the social role (Leslie, 2015, p. 130).

This normative, imperative-like, force of generics about social identity groups affects the nature of meta-linguistic stereotypes about them. This can be captured by expanding the notion of context to include not only propositions in the common ground, but also actions on to-do lists (Portner, 2004, 2007). Very roughly, the to-do list function assigns properties onto each interlocutor’s to-do list which represents the actions that they are expected to perform. The normative dimension of social identity stereotypes can be understood as the acts that are considered appropriate or even obligatory for a member of that group to perform. As such, in a context where the interlocutors accept the social identity stereotype that boys do not cry, the property of not crying will be assigned by default to the to-do list of boys that enter that conversational context. Since social identity stereotypes include items on to-do lists, they affect how interlocutors expect others to behave based on their social identities. If an individual acts in a way that diverges from these expectations, s/he will be considered to act inappropriately. For instance, if the context includes the stereotype that women are submissive, then being submissive will be assigned by default onto female interlocutors’ to-do lists.

Meta-linguistic stereotypes provide an important tool in giving a structural account of implicit bias since they constitute shared sociolinguistic dispositions rather than individual mental states. Hence, testimonial injustice does not only arise when the listener believes a prejudiced stereotype, but it also when the interlocutors accept such stereotypes in the discourse context – even if neither of them explicitly believes it.