Language and Prejudice

Explicit Expressions of Prejudice

The simplest way to conceptualize the effect of prejudice on language is the explicit or overt expression of prejudiced thoughts and beliefs, such as hate speech and derogatory slurs. Very few studies have, in recent years, studied explicit effects. Those that have are concerned with either (a) the features of language such as word choice, phrasing, and style, which are used strategically to persuade and incite prejudice in others (Taylor, 2007) or (b) the relationship between explicit expressions of prejudice to measures of prejudice and behavior (Poteat & DiGiovanni, 2010). In all cases, there is malicious intent since language is seen as a vessel that is intentionally used to convey prejudice beliefs. Today, however, explicit expressions of prejudice are relatively rare given current social norms condemning them, which might explain the lack of research in this field.

Implicit Expressions of Prejudice

In contrast to the explicit expression of prejudice, our language can be affected by prejudice in more subtle ways. Prejudice, just like any belief that we hold, can, without our awareness, influence the words we choose to use. Thus, language can reflect implicit beliefs. Phenomena of this type have been termed linguistic bias and typically involve the description of stereotype-consistent and stereotype-inconsistent behaviors. The Stereotype Consistency Bias is defined and measured by the amount and type of information that is transmitted. It describes the tendency for more stereotype-consistent than stereotype-inconsistent information to be transmitted. The Linguistic Expectancy Bias (LEB; Wigboldus, Semin, & Spears, 2000) and the Linguistic Intergroup Bias (LIB; Maass et al., 1989) are defined through the use of linguistic abstraction.

Linguistic abstraction is measured using the Linguistic Categories Model (LCM; Semin & Fiedler, 1988), which proposes that a specific behavioral event can be accurately described by any one of four types of increasingly abstract words: Descriptive Action Verbs (DAVs), Interpretive Action Verbs (IAVs), State Verbs (SVs), and Adjectives (ADJs). Thus, if person A hits person B, the event can be described as “A hits B” (DAVs), “A hurts B” (IAVs), “A hates B” (SVs), or “A is aggressive” (ADJs). Thus, the LEB (Wigboldus et al., 2000) is the tendency to describe stereotype-consistent behavior more abstractly than stereotype-inconsistent behavior, regardless of whether the behavior is desirable or undesirable. The LIB (Maass, Salvi, Arcuri, & Semin, 1989) describes the tendency for desirable behaviors by in-group members and undesirable behaviors by out-group members to be described more abstractly than undesirable behaviors by in-group members and desirable behaviors by out-group members, which may reflect the fact that stereotypes tend to be negatively distorted for out-groups (Howard & Rothbart, 1980).

Although Sutton (2010) categorizes research on linguistic bias into language as a vessel, it is arguable that this research is more suited to the barometer category. The key issue is that linguistic bias is expressed without the intention to reveal biased beliefs. Researchers must, therefore, infer the nature of the speakers’ beliefs on the basis of the language used. Linguistic bias is thus used as an indicator of the prejudiced beliefs held by the speaker, or metaphorically, as a barometer to reveal the speakers’ cognition (e.g., von Hippel, Sekaquaptewa, & Vargas, 1997).

Researchers who study implicit effects focus their attention on specific features of language and how they serve to perpetuate prejudice by reflecting prejudiced beliefs. For example, if stereotype-inconsistent information must be communicated, then speakers will unintentionally use subtle features of language, such as negations, that imply that the information is unexpected or atypical (Beukeboom, Finkenauer, & Wigboldus, 2010). If a group member from a stereotypically unintelligent group performs a studious behavior, it is more likely that the behavior will be described using a negation, such as [s/he] did not write a bad report, rather than an affirmation, such as [s/he] did write a good report. These subtle features are of great consequence since they influence the cognition of those exposed to them, reinforcing or inculcating prejudiced beliefs in others and leading to the pervasiveness and ubiquity of prejudiced beliefs. This theoretical argument offers a plausible explanation of how prejudiced beliefs about specific groups become shared despite the increase in social norms inhibiting overt expressions of prejudice. It also has implications for any context in which judgments about a person are made on the basis of their behavior, such as admission and hiring decisions.

Language Distorts Perception

Language can also affect prejudice, and it is this ability of language to influence cognition on which researchers of explicit and implicit expressions of prejudice rely. Before conclusions about the consequences of linguistic bias can be made with any certainty, however, research must also demonstrate that language has a corresponding effect on prejudice. Thus, complementarily to research on implicit expressions, a subset of research on the effect of language on prejudice investigates the corresponding process to see what impact different linguistic choices have on those who are exposed to them. Here, language is a lens that distorts recipient’s perceptions to be more similar to the presumed perception of the speaker. Recent findings reinforce those on implicit effects: negated descriptions, for example, mitigate the valence of the behavior and leads recipients to believe that the speaker actually has opposite expectations for the person performing the behavior when compared with affirmations (Beukeboom et al., 2010; see also Geschke, Sassenberg, Ruhrmann, & Sommer, 2010). In other words, describing the performer of a studious behavior as not stupid would lead the recipient to (a) perceive that the behavior is not as positive as when the performer is described as intelligent, (b) believe that the speaker does not expect the performer to engage in studious behaviors, and (c) attribute the behavior to the situation rather than the performer’s disposition. Mere exposure to derogatory comments about a stigmatized group can also spur negative evaluations of group members (Goodman, Schell, Alexander, & Eidelman, 2008), which suggests, similar to the conclusions of Carnaghi and Maass (2007), that there is merit in campaigns arguing for inclusive and positive language.

There is, however, some evidence that linguistic biases may not contribute to the maintenance of stereotypes. Geschke, Sassenberg, Ruhrmann, and Sommer (2007) found that concrete descriptions of stereotype-inconsistent behavior had a stronger effect than abstract descriptions on recipients’ perception. This suggests that explicit evidence might be necessary to support stereotype-inconsistent assertions, and linguistic bias may be a reflection of this need. It may also be that concrete descriptions have a strong impact on perception in the short-term, but that abstract descriptions play a stronger role in the long term. So while it is clear that linguistic biases are consistently and reliably produced, it is not clear what impact biased messages have on recipients or whether linguistic biases do indeed play a role in the perpetuation of prejudice through this mechanism.

Beyond demonstrating that language has a corresponding effect on prejudice, research in this area shows that words themselves, and the connotations they have, may distort the perception of recipients in ways that are not necessarily similar to the perception of the speaker. Most of this research focuses on gender and language. A clear finding is that mental representations of gender are formed through an interaction of semantic information and stereotypes (Gabriel & Gygax, 2008; Gygax, Gabriel, Sarrasin, Oakhill, & Garnham, 2008). Essentially, words, such as role nouns (e.g., housekeeper and carpenter ), which are intended to be gender neutral, in that they do not have a grammatical or denoted gender, actually activate gender stereotypes (Cacciari & Padovani, 2007). Furthermore, masculine forms of words that are intended to be generic are not interpreted that way; instead, they lead to mental representations that exclude females (see also Stout & Dasgupta, 2011; for consequences not related to gender see Cimpian, 2010).

Peripheral structural features of language can also influence perception. Grammar, for example, requires that comparisons have a direction—either x is compared to y or y is compared to x. Whichever group is the basis for comparison (y in the first formulation and x in the second) forms an implicit norm, where differences from that group need to be articulated and explained. Comparing women to men, for example, would result in statements that focus on women, such as “women are emotional,” since men form the implicit norm. Comparing men to women on the other hand, would result in statements that focus on men, such as “men are strong,” since women form the implicit norm. The group that forms the basis for comparison is the norm so it is not necessary to say, for example, that men are not emotional and women are not strong. Bruckmüller and Abele (2010) show that the implicit norm group is perceived as more powerful and of higher status than the other group, and their typical members are perceived as more agentic and less communal. It is a linguistic convention to compare smaller to larger groups, or compare the group that is perceived to be different to the group that is perceived to be “normal.” This is so ingrained that the direction of comparison now has the power to influence perception. These results provide a demonstration of how linguistic practices serve to perpetuate the status quo of power relations between groups and once again confirm that information beyond semantic content is inferred from communication (see also Maass, Pagani, & Berta, 2007).

A final set of studies indicates that the content of a speaker’s message can affect the recipient’s perception of the speaker. For example, speakers who used abstract language to describe positive behavior and concrete language to describe negative behavior were perceived as more likeable than speakers who abstracted negative behavior and situated positive behavior (Douglas & Sutton, 2010; see also Reitsma-van Rooijen, Semin, & van Leeuwen, 2007). Another study found that speakers who expressed positive stereotypes about a group, that Blacks are athletic for example, were perceived as less likeable, more biased, less qualified for a diversity job, and generally perceived more negatively than speakers who did not express a positive stereotype (Czopp, 2008). Thus, it seems that what a speaker says reflects on how that speaker is perceived by recipients.

In sum, this research shows that language cannot be considered a neutral vessel in which to carry or transmit cognition. It has an unyielding transformative effect: changing what it carries and distorting the perception of those who are exposed to it. This influences prejudice by activating culturally shared ideas and creating implicit expectations that subtly transform the perception of groups and group members.

Language Signals Group Membership

Other aspects of language, beyond the properties of words themselves, can also influence perception. Specifically, peripheral nonstructural aspects of language, such as the accent with which it is spoken, can influence perception by allowing the recipient to make inferences about the speaker’s group membership, suggesting a link between language and social identity. Here, as in the modification to Sutton’s (2010) model proposed earlier, language is a sign; it is not actually distorting perception, cognition, or memory, it is merely making group membership salient, which, as an indirect consequence, may make the speaker a target of prejudice.

The majority of studies show that when language signals group membership, either through accent or the use of in-group-specific phrasing (Mange, Lepastourel, & Georget, 2009), there is a shift in how the speaker is perceived. The central and foremost finding is that speakers who have an accent, or more accurately, speakers who do not have the same accent as the recipients, are evaluated more negatively by recipients than speakers who are perceived not to have an accent (because they have the same accent as recipients; e.g., Hosoda, Stone-Romero, & Walter, 2007). The finding is robust and, in recent years alone, has been replicated with English and Welsh native speakers, with Asian-, regional American-, Spanish-, Latin American-, French-, and Japanese-accented English speakers, and with mild to heavily accented speakers. Native speakers (or speakers who have the same accent as the recipients) are perceived as more potent, intelligent, perceptive, knowledgeable, capable, educated, successful, as a more representative of the in-group, and as having a better ability to communicate than nonnative speakers (e.g., Weyant, 2007). Even the content of language spoken with an accent is subjected to bias (Lev-Ari & Keysar, 2010).

The perception of nonnative speakers is influenced by stereotypes, such that people with some types of accents are perceived as more foreign than others (Cargile, Maeda, Rodriquez, & Rich, 2010) and as more suitable for certain types of jobs (Hosoda & Stone-Romero, 2010). Having an accent is related to fewer feelings of belonging, more perceived stigmatization (Gluszek & Dovidio, 2010), and more depressive symptoms (Kim, Wang, Deng, Alvarez, & Li, 2011). Accents alert the recipient that the speaker is not a part of the in-group and allow the speaker to be a target for prejudices directed toward that group. Research suggests that perspective taking (Weyant, 2007) and awareness of bias (Lev-Ari & Keysar, 2010) might be effective means of reducing prejudice against nonnative speakers. Antiprejudice campaigns should be educational in nature, making people aware of how nonnative speech may indirectly affect their perception and encouraging them to consider the nonnative speaker’s perspective.

Using language as a sign of group membership might have been evolutionarily advantageous because even infants and children show a preference for native speakers (Kinzler, Corriveau, & Harris, 2010; Kinzler, Dupoux, & Spelke, 2007). Since this preference develops earlier than racial preferences (Kinzler & Spelke, 2011), language may have been a more useful signal than race in the past. Indeed, neighboring communities were likely to have differed by regional accent, but not by race, making language a more reliable cue to in-group membership and highlighting the link between language and identity. In support of this, one study suggests that language can signal one’s own group membership. In this study, bilingual participants completed the same Implicit Association Task (IAT) for attitudes toward Moroccan and French people in Arabic and French. Participants had stronger pro-Moroccan attitudes when they completed the IAT in Arabic versus French. In a second study, participants had more pro-Hispanic attitudes when they completed a Hispanic American IAT in Spanish rather than English. In essence, language influenced positive implicit attitudes by shifting them in the direction of the linguistic in-group associated with the language being used (Ogunnaike, Dunham, & Banaji, 2010). This demonstrates an interconnection between language and identity, such that using a specific language activates shared cultural beliefs of, and encourages identification with, the social group associated with it.

Language Defines Group Boundaries

Language can also be used to define the boundaries between groups, emphasizing the link between language and identity. Here, language is either a sign or a lens. The social identity approach suggests that it is intimately connected with the nature of intergroup relations (Tajfel & Turner, 1979; Turner, 1999). Accordingly, research in this area demonstrates that using language as an in-group attribute (perhaps strategically) to define who is a member of the in-group, must also include, by default, a definition of the out-group and, therefore, has consequences for intergroup relations. For example, national identification predicts anti-immigrant prejudice most strongly in countries where national identity is based on language (e.g., Matser et al., 2010; Pehrson, Vignoles, & Brown, 2009). In this case, language is a sign, signalling the in-group identity and, indirectly, affecting intergroup relations. It does not necessarily distort or change perception, but is an attribute of the in-group.

On the other hand, if language is an in-group attribute, then group boundaries may be less definite and more permeable since people are able to learn other languages. Learning a second language then may break down barriers between groups, and a recent study suggests that this is the case (Ekşi, 2009) similar to Wright and Bougie (2007) and Rubenfeld et al. (2007). In this case, language is a lens, transforming the monolingual speaker into a multilingual speaker with multiple and changing identities. In sum, language can play a more central role in the definition of group boundaries, increasing or decreasing the permeability and strength of, or distance between, groups.

Group Membership

The first moderating aspect of the social context is group membership: Speakers do not operate in isolation, where the information they convey is a sole function of intention. First, group membership of the speaker, through stereotypes, can affect how the same words are interpreted by the same group of recipients. For example, Maoz (2009) found that a proposal for compromise offered by a female opponent, in an intergroup negotiation context, is valued more favorably than the same proposal offered by a male opponent. Presumably, stereotypes that women are warmer, more trustworthy, and less aggressive than men inform expectations for, and influence the perception of, an individual woman’s language and how her message is received. Collins, Biernat, and Eidelman (2009) provide a demonstration of how stereotypes can affect interpretation through common ground. When asked to interpret descriptions of academic records back into objective academic scores, translators decoded more favorable descriptions into weaker academic performances when the student was purported to be Black. No explicitly racist comments were expressed but, implicitly, what interlocutors were saying was that “the Black student did well for a Black person (with presumably lower standards).” Through an analysis of language within the appropriate communicative context, language that was not overtly negative (and was in fact ostensibly positive) was shown to be a reflection of a deeply rooted prejudice against Black people. In sum, the group membership of the speaker and the person being discussed influence the interpretation of what is communicated.

Other studies investigate how the group membership of the speaker, recipient, and subject of communication may influence what is and what is not expressed. The major contribution of this research is to place the previously mentioned linguistic biases in context, showing definitively that linguistic bias is not as straightforward as previous research would suggest. In fact, it appears that the LIB is only expressed by high- and low-status groups (Moscatelli, Albarello, & Rubini, 2008) and is used differently by majority and minority group members (Shulman, Collins, & Clément, 2011). Both minority and majority groups, for example, express a LIB and a reverse LIB, but majority members are more likely to show a pronounced LIB effect while minority members are more likely to show a pronounced reverse LIB effect. Essentially, the LIB is or is not expressed based on the group membership of the interlocutors. This contributes to research on prejudice by indicating that status asymmetries and power relations between groups enhance out-group derogation. It also indicates that our language does not always reflect our implicit beliefs and may be subject to influence from the communicative context. More specifically, the group membership of the interlocutors and the status differences between their respective groups will influence whether or not, and how, a linguistic bias is expressed. Klein, Clark, and Lyons (2010) show that there is a greater stereotype consistency bias between friends and that it correlates with measures of interpersonal closeness. This suggests that the stereotype consistency bias might even be a tool that can be used strategically to achieve intimacy with conversational partners.

A final study adds another layer of complexity to the studies on linguistic bias. Participants who were exposed to language that was subtly biased against them perceived more interpersonal distance with their communication partner than participants exposed to language that was biased toward them (Reitsma-van Rooijen et al., 2007). This would, presumably, result in lower levels of communication satisfaction and a resistance to engaging in intergroup interaction in the future. Here, language is a sign of group membership and a tool that can be used to suggest social distance between interlocutors.

Overall, this research brings into question previous formulations about the nature of linguistic bias. If linguistic bias is truly an implicit indicator of prejudice and not subject to conscious control (Franco & Maass, 1996; von Hippel et al., 1997), then it cannot theoretically be used strategically or be subject to influence from the social context as recent research suggests. In sum, the group membership of the speaker, recipient, and subject of conversation, all combine to create a communicative context through which all communication must be interpreted. Implicit biases may or may not be transmitted depending on the intergroup nature of the context, and paradoxically, may even be used intentionally for some communicative goal.

Communication Strategy

The communication strategy of the interlocutors is another aspect of social context that moderates the relationship between language and prejudice, as the content and style of communication may be intentionally altered to meet interlocutors’ communicative goals. In this area, language is conceptualized as a barometer since research goes beyond the face value of the content in communicated messages to try to interpret the speakers’ intentions.

First, results show that the communication strategies a speaker employs during intergroup interactions can change the language that is used or the style with which it is communicated (e.g., Ladegaard, 2011). Second, speakers from different groups pursue different impression management goals, and have different communication strategies, in intergroup interactions, and these are based on the stereotypes of the speaker’s group (e.g., Bergsieker, Shelton, & Richeson, 2010). For example, Whites, who are worried about appearing prejudiced during interracial interactions, strive to appear moral and want to be liked by their conversation partner. Minorities, who are worried about being negatively evaluated, strive to appear competent and want to be respected by their conversation partner. Third, groups also differ in what they believe are the most useful strategies to achieve the same impression management goals. For example, both Black men and Black women are concerned with being negatively evaluated by a White conversational partner but Black women think that being nicer than usual will garner a positive impression while Black men think that adhering to White norms is a more useful strategy (Shapiro & Neuberg, 2008).

This research also demonstrates, however, that the intentional use of communication strategies can be ineffective. For example, women who adopt a more masculine style of communication (e.g., more direct, with fewer hedges and hesitations; von Hippel, Wiryakusuma, Bowden, & Shochet, 2011) to appear as a more effective leader are actually perceived less favorably, and communication partners were less willing to comply with their requests. Furthermore, using a communicative strategy may influence the quality of the intergroup interaction. For example, Black patients perceived White physicians who were high on both implicit and explicit prejudice more favorably than those high on implicit and low on explicit prejudice (Penner et al., 2010). This suggests that conforming to social norms, or having a communicative agenda, such as attempting to appear unprejudiced, forces one to act incongruently with one’s implicit (and, presumably, true) beliefs. This subtly disrupts interaction behaviors, leading to less positive interactions and lower levels of communication satisfaction.

Social Norms

Social norms regulate all public behavior by specifying what behaviors, attitudes, and beliefs are expected and acceptable in all contexts in which we are in contact with others. Language is particularly subject to social norms as they define how appropriate it is to express certain views and beliefs. A recurrent finding is that most statements that are made to an audience are done so in support of implicit norms—that speakers’ messages conform to the speakers’ perception of prevailing views (e.g., Shapiro & Neuberg, 2008; see also Pratto, Korchmaros, & Hegarty, 2007). Research also shows that this may actually change the speakers’ perception of the message content (e.g., Carnaghi & Yzerbyt, 2007). Language, through group discussions, may also change norms or the effect of stereotypes via the creation of new local norms (e.g., Smith & Postmes, 2009, 2011).

Interlocutors are aware of situational constraints on communication but the consequences of this are not clear. Conway et al. (2009) asked participants to read a short story about a group of students discussing a fraternity in a positive manner. In the censorship norms condition, a fraternity member was present during the discussion. This implicitly suggests a censorship norm—that making negative comments about the fraternity would be inappropriate in this situation. When participants were asked what they would communicate about the fraternity to a friend (with no fraternity member present), participants in the censorship norms condition were more likely to disparage the fraternity. This suggests that self-censorship norms can backfire: Participants were aware of the situational constraints on the opinions expressed and overcompensated for those constraints later. A second study suggests, however, that social constraints may temper reactions, leading to more positive intergroup interactions (Mallett & Wagner, 2011). In this study, male participants were confronted by a female confederate for being sexist because they had ostensibly used the pronoun she to refer to a nurse in a discussion of a moral dilemma. After a confrontation, participants compensated for their sexist behavior, experienced more mutual liking with the confederate, and were better at detecting sexist language than participants who did not experience a confrontation. In this case, self-censorship led to more positive interactions and increased the participant’s awareness of sexist language. Thus, it is not clear whether self-censorship norms, such as those advocated by positive and inclusive language campaigns, are effective at reducing prejudice, as research on direct effects would suggest.