Dogwhistles and Audience Design: A New Definition

Abstract In recent years, scholars have vividly debated over the definition and features of dogwhistles. As Jennifer Saul has widely argued in her works, political dogwhistles are powerful tools of manipulation. However, the current debate still lacks a convincing definition of dogwhistles, which sometimes are treated like spy codes while, at other times, they are labelled as instances of hate speech, as in Santana (2019). Instead, I propose a definition of dogwhistles that is based on the analysis of the audience design of utterances. I claim that dogwhistles are speech acts designed to secretly change the conversational role of a subset of the audience. Furthermore, they qualify as forms of disguisement - and not concealment, as claimed by the received view - that violate two important conversational responsibilities of the speaker (Clark and Carlson 1992).


Introduction
There is a growing philosophical and linguistic literature addressing the topic of dogwhistles.The metaphorical use of this term was legitimized for the first time by US political journalism in the late eighties 1 .Since then, common sense and the academic literature have slowly adjusted the political metaphor of dogwhistles, and now this class of expressions refers to a particular kind of manipulation usually perpetrated by politicians and their spin doctors.As Drainville and Saul (forthcoming) state, "the exact definition of 'dogwhistle' is a matter of debate, but -very roughlythese are utterances which function by concealing their meaning or intended effect from at least some of the audience".Instances of dogwhistles in American politics are expressions, such as 'inner-city' or 'illegal immigrants', which convey at the same time two different meanings: a first nonloaded one that is easily grasped by the whole audience and a second more problematic one (because discriminatory) that, instead, is transmitted only to a subset of the same audience.Even if some scholars (Santana 2019) have refused to construe the definition of dogwhistles as matching the working principles of secret coding, as in the case of Saul (2018a), I argue that we can still maintain a definition of dogwhistles that shares some features with secret coding by appealing to the notions of audience design presented by Herbert H. Clark and Thomas B. Carlson in "Hearers and

Defining dogwhistles
Although the debate around the topic is quite new in the philosophical literature, we already have an array of definitions of dogwhistles.The received view on dogwhistles, introduced by Witten (2014) and endorsed by Saul (2018a), is the following: A dogwhistle is a speech act designed, with intent, to allow two plausible interpretations, with one interpretation being a private, coded message targeted for a subset of the general audience, and concealed in such a way that this general audience is unaware of the existence of the second, coded interpretation.(Witten 2014, p. 2) Just like in a non-metaphorical sense there are highfrequency sounds that are perceptible only by dogs, in the same way there exist words, utterances or images capable of communicating different messages in different circumstances, so that different audiences will get different messages.Saul distinguishes between two kinds of dogwhistles: overt ones and covert ones.This is a distinction that looks at the way dogwhistles affect hearers.A dogwhistle is overt when every audience consciously gets the message meant for it.In contrast, a dogwhistle is covert when audiences are not aware of the message transmitted, since it is delivered thanks to exploitation of hearers' bias and prejudices.In turn, every dogwhistle can be intentional or unintentional.This time it is a distinction that pertains to dogwhistle-utterers.A dogwhistle is intentional when the speaker is aware of the effect that might be produced by her words on the audience, and she utters them in order to produce it.In contrast, it is a case of unintentional dogwhistle when the speaker is not aware of the possible effect of her words and so she utters them without the intent to produce it.By merging every characterization we end up with four combinations: overt intentional dogwhistles, covert intentional dogwhistles, overt unintentional dogwhistles and covert unintentional dogwhistles.Even if covert dogwhistles, as well as unintentional ones, represent a very interesting phenomenon, in this paper we will deal only with overt intentional dogwhistles.
Other scholars like Henderson and McCready (2019) propose a definition that is fairly similar to the one endorsed by Saul.According to them, dogwhistles is language that sends one message to an outgroup while at the same time sending a second (often taboo, controversial, or inflammatory) message to an ingroup.(Henderson and McCready 2019, p. 223) This second definition adds another facet of dogwhistles, namely their discriminatory dimension.Indeed the second, concealed message of dogwhistles is often used to convey racist, sexist, or other prejudiced content that is targeted to social groups (usually discriminated groups).Consider the example of inner-city in the phrase "We are going to crack down inner-city violence".As Jason Stanley (2015) argued, code words2 like 'inner-city' have a conventional meaning that, in this case, roughly amounts to something like "the central part of the town".On the basis of specific literature in linguistics (see Beaver et al. 2009, Simons et al. 2010, Murray 2014), Stanley analyses this primary meaning as atissue content: it is content that is subjected to the negotiation of interlocutors before entering the common ground of the conversation (Murray 2014, Stanley 2015).It is content that is directly relevant to the conversation at hand.But at the same time, Stanley claims, terms like 'inner-city' also have a secondary, often discriminatory, meaning -what he indicates as (and linguists call) not-at-issue content.Not-at-issue content is content that enters the common ground without being subjected to any kind of negotiation from the interlocutors: "The not-at-issue content of an utterance is not advanced as a proposal of a content to be added to the common ground.Not-at-issue content is directly added to the common ground" (Stanley 2015, p. 135).It is content somehow backgrounded, because it does not in itself move the conversation forward in its established direction.In the case of 'inner-city', the notat-issue content is racist content against Afro-Americans that sounds like "Black people are violent/aggressive".
Leaving aside Stanley's framework, the discriminatory dimension of dogwhistle is widely acknowledged in the literature.Carlos Santana (2019), for instance, criticized the received view on dogwhistles and renamed it the secret code account because of the alleged similarity between dogwhistles (as interpreted by this view) and the functioning of coded meanings like spy codes3 .Santana argues that the secret code account cannot be maintained because the alleged concealed meaning is often recovered by the general audience.People are usually aware of the discriminatory meaning conveyed by 'inner-city' or 'illegal immigrants', so there is no reason for basing the definition of dogwhistles on the concealment feature.Santana, instead, proposes a different definition of dogwhistles that revolves around their strategic utility.According to Santana, an act of political communication is a dogwhistle to the extent that: a.It has a secondary, implicit meaning in addition to its surface meaning.b.The secondary meaning, but not the surface meaning, calls attention to a politically meaningful social category in a way that violates widely shared egalitarian norms.-Rev. Int. Fil. Campinas, v.46, n.3, e-2022-0071-R3.c.A certain audience approves of violating these norms, while another different audience prefers adherence to them.d.That norm violation can be plausibly denied by the dogwhistle user.

Manuscrito
The originality of Santana's definition lies in d: the primary reason a speaker uses a dogwhistle is to obtain plausible deniability4 .For this very reason, let's call his account the deniability account of dogwhistles.Consider the use of the term 'welfare queen' in Ronald Reagan election campaign.According to Santana, the surface meaning of welfare queen is "woman who abuses the welfare system", while its secondary meaning is "welfare is a system that supports lazy, greedy, undeserving black women".Stating the secondary meaning out loud would cause political problems to the speaker, whose reputation might be irreversibly damaged by racist accusations.Instead, dogwhistles like welfare queen enables Reagan to successfully deny that he intended something racist, "since it's hard to hold someone accountable for the implicit content of their statements" (Santana 2019, p. 6).Thus, obtaining plausible deniability is what distinguishes dogwhistles from explicitly homophobic, sexist and racist utterances.Furthermore, Santana also argues that we should put at least some dogwhistles under the wider umbrella of hate speech and/or slurs.That is because, for instance, "welfare queen" is a stereotypical description of black women and counts like a slur in a descriptive account of slurs (Williamson 2009, Hom 2010), and so on.
I'll reply to Santana's criticisms of the secret code account in the last section, after putting forth a new account of dogwhistles.To this end, let me provide some preliminary remarks about audience design and speaker's responsibilities towards any audience subdivision.

Audience design and speaker's responsibilities
In conversation, speakers exploit knowledge, assumptions, beliefs they take to be shared with the addressees in order to communicate something.Normally, we call this set of knowledge, assumptions, beliefs mutually shared by speakers and addressees the common ground of the conversation 5 .The common ground of a conversation is a dynamic and unstable entity: it can be updated in the course of the same conversation by adding, removing and adjusting the 'material' inside it.Thanks to this feature, the speaker shapes her utterances on the basis of the more updated version of the common ground and, thus, makes sure (most of the time) the uptake of the addressee.This entails that the speaker must adjust her utterances in the event that contextual changes should occur.Therefore, the speaker bears some conversational responsibilities towards the addressee.But the same goes for the other party as well: the addressee, for instance, is responsible for keeping track of what the speaker is saying to her.Thus, in general, the following principle applies:  Principle of Responsibility.In a conversation, every party is responsible for keeping track of what is said, and for enabling the other parties to keep track of what is said6 .
The Principle of Responsibility establishes the primary conversational responsibility among parties involved in a conversation: keeping track of what is said and enabling others to do the same.Tracked information of what is said is stored in a subset of the common ground called "discourse record".That is why, in addition to sheer propositional content like beliefs and assumptions, we must also cite previous conversational moves as part and parcel of the common ground.However, this two-party model of the conversation is not really in line with real occurrences of linguistic exchanges.In fact, the category of listeners in a conversation amounts to more than just speakers and addressees.According to Clark and Schaefer (1992), conversational roles should be divided in three main categories:  Speaker.She can be considered as a part of the listeners too because she listens to herself in order to monitor what she is saying (but we will leave aside this role of monitoring). Participants.These are the listeners who the speakers intend to actively include in the conversation.By 'actively include' I mean that the speaker bears reflexive or Searlean intentions towards them.An intention is a reflexive one iff "the speaker S intends to produce an illocutionary effect IE in the hearer H by means of getting H to recognize S's intention to produce IE" (Searle 1969, p. 47)  However, the second and the third macro-category need at least two further internal subdivisions (Clark and Schaefer 1992).Among participants we must distinguish the 7 The difference between reflexive intentions and not reflexive ones is almost the same as that between communicative and informative ones in Grice.Here, though, the definition of reflexive intention incorporates the notion of illocutionary effect as we find it in Searle (1969).Unlike Grice (1957), Searle's definition does not hinge on the production of some general "effect" in the hearer.Thus, the Searlean intention consists simply in the hearer understanding the utterance of the speaker: it does not need the generation of a new belief or a response on the part of the hearer.One of the criticisms put forth by Searle against Grice's nonnatural meaning, in fact, is that it confuses illocutionary acts with perlocutionary ones.addressee(s) of the utterance -the ostensible target of what is said -from the others, which are called the side participants.Side participants might become, in turn, addressees or speakers at a later time.Imagine that Giacomo, a friend of mine, is present when I ask Paolo to close the window: while Paolo is clearly the addressee of my illocutionary act, Giacomo is only a side participant (although the bare presence on its own is an insufficient criterion to determine who is a side participant, see footnote 11 below and its related discussion).At the same time, we should distinguish two kinds of overhearers: 1) the bystanders, listeners who have access to what the speaker is saying and whose presence is fully recognized by the speaker herself; 2) the eavesdroppers, listeners who have access to what the speaker is saying but whose presence is not fully recognized by the speaker herself.With "fully recognized by the speaker" Clark and Schaefer (1992) intend that speakers believe that they and the bystanders mutually believe that they, the bystanders, have access to what is going on (they can listen to the conversation, for instance), even if they are not actively included in the conversation in our technical sense.In contrast, speakers believe that they and the eavesdroppers, if there are any, do not have this mutual belief 8 .In the example made before, Maria is an eavesdropper: since she is in the next room, her presence is not fully recognized by the speaker and she is not the recipient of his reflexive intentions.As an example of bystander, we can imagine two friends talking to each other within a crowded bus and surrounded by strangers.In that case, every stranger represents a bystander for their conversation, because they are not actively included in the conversation but still their presence is fully recognized by the speaker 'in charge'.But how does this role assignment occur?As Clarke and Carlson write: When the speakers design their utterances, they assign different hearers to different roles; and then decide how to say what they say on the basis of what they know, believe, and suppose that these hearers, in their assigned roles, know, believe, and suppose.(Clarke and Carlson 1982, p. 217) Clarke and Carlson's idea is that roles are completely determined by the speaker with the audience design of their utterances.'Audience design' refers to a set of utterance's features that the speaker deploys in order to make the role assignment.Some are external features.External features are something like the physical arrangement (for instance, participants are usually near each other), the history of previous conversational moves (if you are a participant during the last utterance, and the speaker gives no indication of sudden changes, you can take for granted that you are also a participant for the current utterance), the gestures, the manners of speaking.Other features are internal to the utterance.Addresses, for example, can be elected by means of vocatives, while participants can be taken in the conversation by prefatory utterances 9 .Regardless of which feature is the prominent one, it is essential that the audience design enables both the speaker and the addressee(s), and the speaker and side participants, to be mutually aware of which listeners are being designated as addressee(s) and which as side participants.Keep in mind that the role assignment of the audience design is not established once and for all: speakers assign these roles for only limited periods of time.Even if Paolo is my addressee in the first place, I can switch his role with Giacomo's with my next utterance; and I can do it in the midst of the same utterance and multiple times per utterance.But if the role assignment depends on the audience design of the utterances, then the fulfilment of the mutually awareness condition is the speaker's conversational responsibility -at least in part.
However, the audience design of speaker's utterances takes into account the category of overhearers too.Since overhearers are not actively included in the conversation by definition, speakers do not bear primary conversational responsibilities towards them, according to Clark and Schaefer (1992).When Anna and Giacomo are talking to each other within a crowded bus, they do not strive to enable strangers to keep track of what is said.That is because the kind of intentions that the speakers has towards overhearers are not intended to be recognized as intended to be recognized (i.e. they are not reflexive).Most likely, depending on the degree of secrecy around the topic, Anna and Giacomo may try to conceal the content of their conversation by speaking softly or using obscure referential expressions like 'you-know-who' and such.Concerning overhearers, speakers may still bear secondary conversational responsibilities towards them, like politeness responsibilities: they must avoid threatening overhearers' face without reason (see Brown and Levinson 1987) 10 .This is true for both bystanders and eavesdroppers although here we won't deal with politeness responsibilities.
The term 'overhearer' also includes eavesdroppers since the speaker may weave utterances by considering unknown overhearers too.However, even if this scenario is perfectly conceivable, I think that we should focus only on bystanders.The reason is that the role of bystanders, contrary to eavesdroppers, is fully assigned by means of the audience design of speaker's utterance.Although the speaker can design her utterances with eavesdroppers in mind, she never has the same degree of awareness as overhearers whose presence is fully recognized -that is, bystanders.This fact gives the speaker the possibility to adjust her audience design accordingly to determine who is a bystander for the next stretch of the conversation.Consider the bus example again.The use of obscure referential expressions as internal features of audience design does not only serve the purpose of concealing the message from the strangers; most importantly, the same act of concealment also serves the purpose of assigning the role of bystanders to the surrounding strangers.That is why the bystander is the one who is present but at the same time is not actively included in the conversation.In summary, the speaker has always indirect responsibility, implicit in the audience design, to let someone know that she is a bystander.Again, speakers assign listeners to their conversational roles by the way they engineer internal and external features of their utterances in the current situation.Therefore, "speakers must get listeners to recognize their assigned roles" (Clark and Schaefer 1992, p. 250).Their responsibility, here, is not exactly to tell people that they are bystanders, but to use an audience design that accomplishes this goal 11 .This kind of responsibility is only 11 That is why the mere physical presence cannot distinguish a side participant from a bystander: it is always audience design that enable speakers to determine the role assignment.Since audience design is the means through which speakers attempt to fulfil their reflexive intentions, we must look at it to determine who is a side limited to the reception of one's role assignment within conversation: it has nothing to do with a different (but potentially overlapping) speaker's moral responsibility about, say, potential messages that she may convey to nearby overhearers for some reason, e.g.making the other person overhear something on purpose to manipulate their behaviour 12 .It is only a matter of conversational responsibilities.But the same responsibility does not hold likewise for listeners whose presence is not fully recognized by the speaker, because she cannot easily envision potential overhearers and because, above all, eavesdroppers do not recover their role from the audience design of speaker's utterance: it is not the speaker that assigns their role, for speakers are not aware of their presence by definition and so they cannot adjust their utterances accordingly.If anything, the speaker can only design her utterances so that politeness responsibilities are maintained towards potential eavesdroppers too.This last kind of speaker's responsibility participant and who is not.For instance, external features of audience design can contain cues about it.Repetitive gestures of the speaker directed at the same person are a clear indication of active inclusion in the conversation.Furthermore, the speaker can maintain eye contact with side participants while she is addressing someone else, thus signalling that they are still actively included in the conversation and might also become addressee in the next stretch of the conversation.That is how side participants can be distinguished from bystanders in the crowded bus situation.However, this does not mean that audience design is always infallible.We can easily imagine cases in which the speaker does everything possible to get listeners to recognize their assigned roles by means of her audience design, and yet she fails anyway.But speakers have the responsibility to engineer an audience design that is at least conducive to that goal (thanks to the anonymous reviewer for their remark on this point).
-the indirect one -is one of the cornerstones of the dogwhistles' puzzle, as I would like to argue in the next paragraphs.First, however, I need to briefly address two ways of dealing with overhearers: concealment and disguisement.

Concealment vs disguisement
We already said that audience design also takes into account overhearers.As Clark and Carlson say: Speakers also design their utterances with overhearers in mind.Although they don't intend the overhearers to 'take part in' what they are saying -in the favoured sense of 'take part in [namely, bearing reflexive intentions towards them] -they realize that the overhearers can nevertheless form conjectures or hypotheses about what they mean.The purpose of overhearer design is to deal with these hypotheses.By designing their utterances just right, speakers can lead overhearers to form correct hypotheses, incorrect hypotheses, or even no coherent hypotheses at all.(Clark and Carlson 1982, pp. 220-221) Since they do not bear primary conversational responsibility towards overhearers, according to Clark and Carlson (1982) speakers can maintain one of the following four attitudes towards this audience category: indifference, disclosure, concealment, disguisement.The first and the second attitude are of little interest to us.Suffice it to say that, while indifference means that the speaker does not care about whether or not the overhearers can grasp what she is saying, disclosure entails designing the utterance so that overhearers can grasp it fully.Concealment and disguisement are much more interesting for our purposes.
Let's go back to the window's example.Imagine that all the listeners, including Maria in the next room, are Italian native speakers.In this case, when I ask Paolo to close the window (in Italian) the common ground includes information about our shared language that Maria can easily grasp.Open common ground information is information that the overhearer can easily conjecture to be in the common ground shared by participants.Now, instead, imagine I suddenly point out the same window saying "You saw the condition it is in" to Giacomo.In this case, I am exploiting a joint perceptual experience between me and Giacomo in order to focus his attention on visual details; but now, since Maria is in the next room (i.e.she does not share the same portion of common ground in terms of perceptual experiences), she cannot readily guess the piece of information that I am trying to get across.This is a clear example of closed common ground information: information that the overhearer cannot easily conjecture to be in the common ground among participants.Closed information is at the base of concealment attitude: Concealment.For any chunk of the intended meaning, the speaker designs her utterances so that overhearers cannot grasp it and will recognize that they cannot do so 13 .
It is based on the exploitation of closed common ground information: the speakers design their utterance on the basis of the shared common ground with participants, but on parts of the same common ground which are thought to be closed to the overhear.If I am aware of the presence of someone else in the next room who is not supposed to be actively included in the conversation (= a bystander), and I want to conceal my conversation, maybe I can start speaking in Japanese with other participants, thus exploiting a (supposed) closed part of the common ground among participants 14 .The use of obscure referential expressions in the crowded bus is another instance of concealment.I think that this is the very sense by which the received view on dogwhistles explains their functioning.The secret code account of dogwhistles, in my view, is an account of concealment: the speaker exploits closed common ground information shared with a subset of the general audience to deliver a secret, often inflammatory message that is not captured by the rest of the audience.And what about disguisement?Clark and Schaefer (1992) describe the attitude of disguisement as follows: Disguisement.For any part of what they mean, speakers can design their utterances so that overhearers will be deceived into thinking it is something that it is not.
As Clark and Schaefer rightly recognise, with concealment overhearers (bystanders) normally notice that they are being kept in the dark.When Anna and Giacomo switch to obscure referential expressions, nearby strangers usually realize that they are being treated as overhearers and 14 It only depends on my knowledge of Maria's personal identity (and so on the common ground between me and her).If she is a complete stranger to me, then concealment is probably harder: perhaps I do not know that Maria is a Japanese native speaker too and I would fare better by using a vocabulary only Giacomo and metha understand, for instance.so they are not actively included in the conversation.But with the attitude of disguisement overhearers do not fully recognize that they cannot grasp the intended meaning.Disguisement is the most sophisticated attitude inasmuch as it entails misrepresenting, at the expense of the overhearer, what is open common ground information and what is closed information.A simple case of disguisement is switching to a language that masquerades as a genuine one, making the overhearers believing that they are having access to open information.One historical example may be the World War II conversation between Kurusu, the Japanese Ambassador to the US, and Yamamoto, the Foreign Office American Division Chief, before Pearl Harbor: But without anything, they want to keep carrying on the matrimonial question.They do.In the meantime we're faced with the excitement of having a child born.On top of that Tokugawa is really champing at the bit, isn't he? (Clark and Carlson 1982, pp. 221-222) In that conversation Kurusu was reporting that Japan's military action against the US was imminent, but the ambassador deployed a language in disguise in order to escape telephone tapping: 'matrimonial question' secretly referred to the negotiations, 'having a child born' was an alternative expression for 'crisis' and 'Tokugawa' was a reference for the Japanese army.Note that it is not the same kind of code that pertains to concealment and the secret code account.In fact, the latter considers only codes that enable unwanted listeners to be recognized as such.But if you were in charge of intercepting Kurusu's conversation, it is very likely that you would not perceive the artificial ambiguity of 'matrimonial question'.Therefore you might erroneously think that you have full access to the common ground of the conversation, while in fact you are a victim of disguisement.What makes disguisement so demanding is that the speaker must be capable of simultaneously doing three things: 1) to get the addressee and side participants, if any, to recognize her intended meaning; 2) to conceal the same meaning from overhearers and 3) to get overhearers to think that she means something else (that is related to open common ground information).In some circumstances, the task might be facilitated; this occurs when, for instance, the speaker knows the identity of overhears and so shares with them at least some salient parts of the common ground (and by 'salient' I mean parts that are useful to carry out the disguisement).In contrast, as Clark and Schaefer also suggest, "it is hard enough to disguise parts of what is meant -such as references to people, places or objects -and to disguise hints and other indirect speech acts" (Clark and Schaefer 1992, p. 268).In summary, disguisement is what they brilliantly define as the disclosure of a misrepresentation: the speaker provides overhearers with evidence so that the latter thinks she has understood speaker's utterance when it is not true15 .Now, we finally have everything we need to define dogwhistles in a more fine-grained way.

Dogwhistles as forms of disguisement
I refine the definition of (overt intentional) dogwhistles as follows: Dogwhistle.A speech act designed to change the conversational role of a subset of the audience from participant to overhearer without making it public knowledge.
My proposal is to consider a dogwhistle as an instance of the disguisement attitude whose effect is twofold.First, it entails a direct violation of the Principle of Responsibility: the speaker fails to meet her primary conversational responsibility towards participants.Consider the 'wonderworking power' dogwhistle in Saul (2018a).In his 2003 State of the Union Speech George W. Bush, according to Saul, was desperately trying to obtain the consent of fundamentalist Christians.His solution was to dogwhistle to them with utterances like this one: (1) Yet there's power, wonder-working power in the goodness and idealism and faith of the American people.
According to Saul's explanation, to non-fundamentalist participants (1) might appear as "an ordinary piece of fluffy political boilerplate, which passes without notice" (Saul 2018a, p. 362).Not so for fundamentalist Christians, who instead will hear the dogwhistle.Within fundamentalists' idiolect, in fact, 'wonder-working power' is an expression that means something like "the power of Christ".Thus, the first coded interpretation of (1) is the following: -Yet there's power, the power of Christ in the goodness and idealism and faith of the American people.
Alternatively, the dogwhistle may simply suggest that George W. Bush is speaking in their idiolect, thus showing that he is an ingroup member.As Saul clearly expresses, the first interpretation Manuscrito -Rev. Int. Fil. Campinas, v.46, n.3, e-2022-0071-R3.is very clearly an overt intentional dogwhistle: it is a coded, concealed message, intended for just a subgroup of the general audience.In fact, it functions rather like the exploitation of a little-known ambiguity.The second is a little messier.It is somewhat like speaking in a regional accent that gives a feeling of kinship to a particular audience.(Saul 2018a, p. 363) Regardless of which message is delivered to fundamentalist Christians, the outcome is the same: a subset of the audience, namely the non-fundamentalist one, will get only the superficial meaning of (1), while the complementary set will also get the private meaning.But what happened from the point view of the conversational roles?
We already said in §2 that conversational roles can easily change in the midst of the conversation.Like in the window example, after asking Paolo I might suddenly turn to Giacomo and start considering the latter as my new addressee.In the 'wonder-working power' example something very similar seems to happen with (1).As soon as George W. Bush utters (1) a role change occurs: the set of participants boils down to fundamentalist Christians (since they become the addressees of the utterance), thus excluding the rest of the audience.However, there are two essential differences compared to the window example.First, the non-fundamentalist set of the audience is turned into an overhearer, and not into a side participant.Nonfundamentalist participants are not intended to get the message of (1) by recognizing Bush's informative intention; in contrast, the former president of the United States does not want, with (1), to actively include them in that stretch of the conversation.Second, and most importantly, George W. Bush does not inform every party about sudden role change: non-fundamentalist hearers are not aware of what is going on.Therefore, in the case of dogwhistle the speaker does not enable all parties to keep track of what is said.The role change is not recorded in the common ground as a new conversational move.This feature is captured by the last part of dogwhistle's definition that states "without making it public knowledge".That is why non-fundamentalist participants in the example above cannot perceive the role change and, consequently, they do not receive the private message.In fact, what the secret code account really accounts for, in my view, is the dogwhistles' affinity with the attitude of disguisement.The concealment of the coded message "in such a way that this general audience is unaware of the existence of the second, coded interpretation" -as the received view maintains -is actually an instance of disguisement.For the truth is that there is no second coded meaning at all.The speaker is only exploiting closed common ground information that is not accessible to the non-fundamentalists, thus treating them as a group of overhearers (bystanders) 16 .
When I say that there is no second coded meaning I am recalling the observations on the issue made by Justin Khoo (2017).According to Khoo, the functioning of dogwhistleterms might be explained by a simple inferential theory that does not postulate any semantically encoded additional meaning (like in Stanley's account, for instance).Khoo invites us to consider an utterance like this: 16 It might be useful to say that in this kind of conversationnamely, public discourses such as speech broadcast on state television -every listener is a potential participant inasmuch as the discourse is recorded and accessible for everyone over time.That is why, in this case, there can be no such a thing like eavesdroppers: the speaker considers the possibility that her discourse may be listened by unknown and unrecognized parties in the course of time (even by future listeners).-Rev. Int. Fil. Campinas, v.46, n.3, e-2022-0071-R3.(A) "The food stamp program will primarily benefit innercity Americans".

Manuscrito
'Food stamp' is a code word that should introduce the thought that Black Americans are lazy because they benefit from social spending programs (Stanley 2015).In Khoo's view, the utterer of (A) did not say anything about African Americans, and she may or may have not conversationally implicated that food stamp programs are primarily directed at poor African American families.However, at least a part of the hearers (here broadly considered) of (A) may also have a preexisting belief like this: (B) "Inner-city neighbourhoods are mostly populated by poor African Americans".
If hearers do believe this, Khoo argues, then they will gather the following inference from the utterance of (A): (C) "The food stamp program will primarily benefit poor African Americans".
It is only the combination of the pre-existing belief (B) together with the utterance of (A) that elicits (C) in the audience: there is no need to assume an ambiguity or a second meaning of "inner-city", as well as there is no need for the speaker to implicitly communicate that the food stamp program will primarily benefit poor African Americans for her speech to have the same effects in her audience (see Khoo 2017 for an extensive discussion).Likewise, in my view, the dogwhistler is appealing to preexisting beliefs within the shared partition of common ground of the targeted audience.In the wonder-working power example, most likely, George W. Bush is exploiting a part of the common ground -shared with fundamentalist Christians -that is relative to religion and its metaphors.For instance, by uttering (1) he may be exploiting the existence of the following belief in their common ground: (B1): "Fundamentalists usually use 'wonder-working power' in the place of 'the power of Christ'".
And then, the same part of the audience may derive the subsequent inference: (C1): "George W. Bush is a Christian" or "George W. Bush is one of us".But, at the same time, George W. Bush pretends to treat non-fundamentalists as participants too by means of the different beliefs related to 'wonder-working power' which are stored in their common ground.Conventional nonfundamentalist beliefs around the meaning of this expression act as open common ground information that the speaker discloses (in our technical sense of 'disclosure', p. 10) to the unwitting overhearers, thus pretending to be overt with them.On this matter, I must stress that my account distinguishes between the illocutionary dimension of the speech act of dogwhistling from its perlocutionary one.Since we have adopted a Searlean account of reflexive intentions, the dogwhistler succeeds in securing the uptake by getting her targeted audience 17 , and only that one, recognize that she is telling them something and what it is she is trying to tell them -the 'private' meaning of the speaker's utterance.In the case of George W. Bush, the private meaning coincides with the first coded interpretation of (1) offered by Saul: "Yet there's power, the power of Christ in the goodness and idealism and faith of the American people".Thus, the illocutionary effect is simply the understanding of the private meaning (again, there is no such a thing as private/second/additional meaning, but it is just the presence of a different set of beliefs in the targeted audience that leads to this second interpretation of (1)).Whereas the subsequent generation of a new inference, like (C1), is a perlocutionary effect that may or may not be fulfilled.Of course, it might be true that in the vast majority of cases the speaker has the generation of a further inference as the invited response of her dogwhistle (what Austin calls the "perlocutionary object" of an illocutionary act), but it is something that must be kept separated from the realization of the illocutionary effect.Now, someone might object that dogwhistlers are not changing the status of a subset of the audience.Instead, since they pretend to treat a portion of the audience as participants too, they have actually created a new audience -the target audience -relative to which the others are just bystanders.However, this is not the case.Consider again the crowded bus example.This time, in addition to Anna and Giacomo, there is a side participant: Carlo.Anna is talking to Giacomo and Carlo, but at some moment Anna suddenly says "Laura called me yesterday".Carlo thinks that Laura is a common acquaintance, but Giacomo knows that by 'Laura' Anna meant Erica, and so Anna has actually communicated to Giacomo only that Erica called her yesterday, while Carlo has become a bystander for that stretch of the conversation 18 .The very fact that Anna pretends to treat Carlo as a participant is the same as saying that his conversational role has actually changed, for the pretense consists precisely in making Carlo believe that he still is a participant when this is no longer the case.The reason is that, at least at some level, Anna is not bearing reflexive intentions towards Carlo.Should the dogwhistle fail because Anna inadvertently ended up addressing and telling Carlo that Erica called her yesterday (e.g.Anna miscalculated portions of common ground shared with Carlo), then this illocutionary effect is not intended to be recognized by means of the speaker's intention to produce it: there was no such reflexive intention on the part of the speaker (as in the case of me and Maria above, albeit Maria is secretly eavesdropping on my statements, so it is not a matter of speaker's miscalculation about shared portions of common ground).Therefore, Carlo is not treated as a participant for that stretch of the conversation: he is actually a bystander 19 .This leads us to the second effect of dogwhistles: the violation of the indirect responsibility to let someone know that she is a bystander.With purely linguistic dogwhistles, this kind of violation is obtained either by adopting an internal feature of the audience design that contradicts external ones or by refraining from using the right internal one.In the case of contradiction, the internal feature is the dogwhistle-term.In the example above it corresponds to 'wonder-working power'.This term contradicts external features in the sense that there is no parallel and coherent change in the manners of speaking, gestures, and in every other external aspect of the audience design deployed by the speaker (Bush).Instead, the external features remain the same as they have been used up to the previous utterance and that had been consistently maintained in the last one.Therefore dogwhistles enable the speaker to maintain the same external features of audience design so that it seems like no role change has really occurred; even if an internal feature -the dogwhistle-termsignals the opposite.In the case of omission, instead, the dogwhistle-term enables the speaker to refrain from using the right tools to address the targeted subset of the audience, which now becomes the unique participant in the conversation (at least until the next utterance).In an ordinary situation, the speaker might have deployed a vocative in order to bring fundamentalist Christians (or any other group) in the conversation as addressees.Obviously, other parties would have found it quite alienating; this is why a dogwhistle is more suitable in these kinds of situation.
As seen above, the 'wonder-working power' dogwhistle observes the three requirements of the disguisement attitude.They get the addressees (the fundamentalists) to recognize their intended meaning (the two possible interpretations).Furthermore, they conceal the same meaning from the overhearer (the non-fundamentalists) and, at the same time, they get overhearers to think that the speaker means something else (i.e.conventional beliefs around 'wonderworking power').And the same goes for the other instances of dogwhistling too.

Comparison with other accounts
In the previous sections I have been putting forth a new definition of dogwhistles that revolves around the secret role change carried out by speakers.However, it is not clear in what sense the proposed account is superior to its competitors so far.Why should we consider dogwhistles as forms of disguisement in our sense?What problems are we able to solve with this new definition at hand?First, I shall recall three salient features of dogwhistles stressed by other accounts: 1. Discriminatory dimension.According to Santana (2019), the alleged secondary meaning of dogwhistle-terms calls attention to a politically meaningful social category in a way that violates widely shared egalitarian norms (that is why Santana regards dogwhistles as slurs or hate speech in disguise).Even if his analysis is more complex, ultimately the same goes for Stanley (2015).A similar remark is made by Henderson and McCready (2019), although the discriminatory dimension in their definition is a matter of typicality and not a necessary condition.2. Double interpretation.Practically almost every account is explicitly or implicitly committed to the idea that dogwhistles are designed to deliver two messages to the audience.Furthermore, according to some accounts (Stanley 2015, Santana 2019) the entire work is done by the supposed secondary meaning of dogwhistle-terms.3. Plausible deniability.Many accounts in the literature acknowledge plausible deniability as a recurrent feature of dogwhistles (Mendelberg 2001, Khoo 2017, Saul 2018, Henderson and McCready 2019, Torices 2021).However, Santana ( 2019) is the only one that builds deniability directly into his definition.
Let us look at them in details.First of all, I believe that the discriminatory dimension should not be considered a necessary condition for dogwhistles, because it excludes cases that we normally regard as such.In other words, it leads to overexclusive definitions.Consider the deniability account, for instance.It locates the peculiarity of dogwhistles in their strategic usage: thanks to the alleged ambiguity of these expressions, the speaker cannot be charged with racism or other assaults against shared egalitarian norms.The attack on shared egalitarian norms is precisely a necessary feature of dogwhistles, according to the same account.However, cases like 'wonder-working power', in my view, suggest that dogwhistles do not necessarily have a discriminatory dimension and cannot be compared to hate speech and/or slurs: the second requirement of the deniability accountnamely, the violation of a shared egalitarian norm -rules out those instances.Santana (2019) does not commit himself to any particular account of hate speech and slurs, but he only mentions the leading accounts on the literature.According to the descriptive account of slurs, for instance, slurs are terms that convey, semantically or pragmatically, discriminatory descriptive content about the slurred against group.Consequently, Santana claims that dogwhistles like "welfare queen" count as slurs in the descriptive account.On the other side, a dogwhistle like "family values" may be used by a politician in order to incite hatred against LGBTQ+ community (by exploiting conservative ideologies, for instance).Yet, in both cases these accounts do not reconcile with dogwhistles like "wonder-working power", in which there is neither manifest conveyance of discriminatory content nor incitement to hatred.Santana's further claim according to which even dogwhistles that function as ingroup handshakes celebrate ingroup membership at the expense of the outgroup seems a little too forced 20 .Instead, my refined definition of dogwhistles just avoids this problem: since it is based on violations of the Principle of Responsibility, my account is not overexclusive towards dogwhistles that do not violate shared egalitarian norms.
Secondly, focusing solely over secret role changes enables us to uniformly include cases of linguistic dogwhistles that may convey more than two messages -like so called "protean" dogwhistles (Saul 2018b) -without creating ad hoc categories 21 .Now, proposals with a clause about double interpretation do not provide for this possibility.This limitation is mainly due to the fact that these accounts tie the 20 "George W. Bush's use of wonder working power as a nod to evangelicals doesn't just say 'I'm evangelical, too' it also says 'and we -unlike Muslims, atheists, and such -are the real Americans'" (Santana 2019, p. 5). 21Saul tells that, during Brexit campaign in UK, the term "immigration" shifted in terms of the target groups that were associated with them, making it difficult to effectively call attention to and criticize the prejudice involved when immigration was invoked as a threat.Therefore, "immigration" was a protean dogwhistle-term insofar as it was designed to allow for more than two plausible interpretations, since the set of target groups included dark-skinned people, Muslims, refugees, Syrians and so on.
functioning of dogwhistles on the alleged double meaning of dogwhistle-terms.Therefore, they end up narrowing the set of 'proper' dogwhistles to the ones that convey at most two messages.Instead, my refined definition of dogwhistles can account for protean cases, for it is not constrained by the number of possible inferences evoked by internal features of audience design: it is sufficient that the speaker secretly downgrades a set of participants as overhearers by means of audience design of her utterance to establish that we are dealing with a dogwhistle.Now, the same audience design may be deployed in different context with different outcomes, so that dogwhistle-terms like "immigration" can sometimes be used to disguise (in our sense) by exploiting closed information about Syrians, or about Muslims in other times and so on.In conclusion, my account may possess a greater theoretical unity overall, inasmuch as it does better than its competitors in gathering different instances of dogwhistles under the same umbrella.
Thirdly I believe, contrary to Santana (2019), that plausible deniability is not a necessary condition for dogwhistles conceived as speech acts.The reason is that plausible deniability cannot be a felicity condition for a speech act22 .Firstly, plausible deniability pertains to speakers, not to speech acts: it is precisely the speaker that obtains/retains plausible deniability.A speaker obtains plausible deniability when a denial becomes available for her as a result of a previous move, be it an insinuation (Camp 2018) or something else.If so, plausible deniability is much more an effect than a pre-condition of speech acts; that is, it belongs to the perlocutionary level of dogwhistles and Manuscrito -Rev. Int. Fil. Campinas, v.46, n.3, e-2022-0071-R3.cannot be considered one of their felicity conditions.A dogwhistle that fails to grant plausible deniability to the speaker does not ipso facto come up against misfire or abuse.It is, instead, a perlocutionary failure inasmuch as the dogwhistle did not produce the intended effect -room for plausible deniability.Therefore, I think it is fair to say that plausible deniability is typical of dogwhistles, provided that we regard it only as a frequent perlocutionary object and not a necessary condition (as for Santana).My definition has the merit of keeping the illocutionary dimension of dogwhistles separated from its perlocutionary one.In my view, a dogwhistle that does not grant plausible deniability can still be a dogwhistle as long as it secretly changes the conversational role of a subset of the audience from participant to overhearer by addressing and producing a specific illocutionary effect (= hearer understanding of the utterance) only in one specific portion of the audience.That said, it is quite understandable why plausible deniability is a recurrent feature in concrete instances of dogwhistling.Disguisement, in fact, is a very risky attitude that demands remarkable skills from the speaker insofar as she must successfully conjecture parts of the common ground are open information for the overhearers and which ones, instead, are closed to them.Furthermore, she must also be capable to misrepresent closed partitions of common ground as open ones in order to deceive overhearers.

Conclusion
I argued that over intentional dogwhistles are speech acts designed to change the conversational role of a subset of the audience, from participant to overhearer, without making it public knowledge.By means of a dogwhistle, the speaker makes a 'cut' on the audience: a subset of the participants becomes addressee, while the rest of the audience is suddenly downgraded to the category of overhearers without its knowledge.This new definition is based on the analysis of the audience design and has some advantages in terms of theoretical unity and inclusion of dogwhistles that do not violate egalitarian norms.This analysis of dogwhistles will hopefully constitute the basis for a further study of the felicity conditions of dogwhistles, which has been neglected so far.