Abstract
This study examined the role of social categorization in the language attitudes process. Participants (N = 1,915) from three ethnolinguistic groups residing in the republic of Georgia—Georgians, Armenians, and Azerbaijanis—listened to a speaker reading a text in a Tbilisi-accented (standard variety) and a Mingrelian-accented (nonstandard variety) Georgian guise. We predicted that the three groups would vary in their ability to correctly categorize the two guises and that this intergroup variation in categorization accuracy would result in intergroup variation in language attitudes. These hypotheses were supported. Georgians were more accurate than Armenians and Azerbaijanis in their categorization of both guises. The Tbilisi-accented (Mingrelian-accented) guise was evaluated more (less) favorably when categorized correctly than when miscategorized. This resulted in intergroup variation in language attitudes: Overall, Georgians evaluated the Tbilisi-accented (Mingrelian-accented) guise more (less) favorably than Armenians and Azerbaijanis, due in part to Georgians’ higher categorization accuracy of both guises.
Variation in language use, such as the use of different accents and dialects, is consequential. Research on language attitudes, or the evaluative reactions to different language varieties, has shown that, worldwide and cross-culturally, people make various attributions about others—including their competence and warmth—simply based on how they speak (Giles & Watson, 2013). The dominant paradigm in this line of research—pioneered by Lambert, Hodgson, Gardner, and Fillenbaum (1960)—requires respondents to listen to audio-recorded voices of a number of speakers—representing at least two contrastive language varieties (e.g., accents, dialects)—and rate each on various traits. Impressions of speakers in these evaluative reaction studies have been theorized to reflect two sequential cognitive processes: social categorization and stereotyping (Lambert et al., 1960; Ryan, 1983). First, listeners use language cues (e.g., accent) to identify speakers’ social group membership(s) (e.g., ethnicity, social class). Second, they utilize those social categories to derive specific stereotypes concerning the speakers.
Although this two-step theoretical model is implicit in most language attitudes research, past studies have rarely asked respondents to identify which social group(s) they think the speakers they are rating belong to and instead have assumed that respondents are categorizing the speakers as intended (i.e., correctly; but see, e.g., Bayard, Weatherall, Gallois, & Pittam, 2001; Lindemann, 2003; McKenzie, 2008, 2015; Yook & Lindemann, 2013). However, at least in some situations, identification may not occur in the manner intended and important individual and group differences in the nature of categorization may emerge, particularly when subtle manipulations of accent and dialect are involved (Preston, 1999; Ryan, 1983). One factor that is likely to influence respondents’ ability to correctly categorize speakers based on language is respondents’ proficiency in the target language. Respondents who lack proficiency in the target language may experience difficulty distinguishing between different varieties of that language and, consequently, may miscategorize speakers of that language (cf. McKenzie, 2008, 2015). The issue of language proficiency is especially pertinent to understanding language attitudes in multilingual settings, where different ethnolinguistic groups’ proficiency in a given language—and thus their ability to distinguish between and correctly identify different varieties of that language—is likely to vary, resulting in intergroup variation in categorization accuracy. Such differences in categorization accuracy are likely to be consequential: Different categorizations activate different stereotypes and, consequently, are likely to engender different evaluations (McKenzie, 2008; Yook & Lindemann, 2013). In this way, intergroup variation in categorization accuracy may produce intergroup variation in language attitudes.
The present large-scale study (N = 1,915) examined intergroup variation in categorization accuracy and its attitudinal consequences. To do this, we systematically examined differences in categorization accuracy and their effects on language attitudes toward two varieties of spoken Georgian—Tbilisi-accented Georgian and Mingrelian-accented Georgian—among three different ethnolinguistic groups residing in the republic of Georgia who vary in their proficiency in the state language—Georgians, Armenians, and Azerbaijanis. In the sections that follow, we first provide a brief overview of the language attitudes literature, focusing on the role that categorization plays in the language attitudes process. Then, we contextualize the present study, describe an experiment designed to test our predictions, and discuss the theoretical, methodological, and practical implications of our findings.
Language Attitudes and the Role of Categorization
As noted at the outset, language attitudes have been theorized to reflect two sequential cognitive processes (Lambert, 1967; Lambert et al., 1960; Ryan, 1983). First, listeners use language cues (e.g., accent) to socially categorize speakers. Second, they attribute to speakers stereotypic traits associated with those (inferred) social categories. In other words, language attitudes have been theorized to reflect people’s stereotypes toward different linguistic groups (see Dragojevic, Giles, & Watson, 2013).
Language-based stereotypes—not unlike other social stereotypes (see Fiske, Cuddy, Glick, & Xu, 2002)—are organized along two primary evaluative dimensions: status (e.g., competent, intelligent) and solidarity (e.g., warm, friendly). Operating within these two dimensions, past research has shown that speakers of standard and nonstandard varieties elicit different evaluative reactions (Giles & Watson, 2013). Standard varieties are those that adhere to codified and accepted norms prescribing “correct” usage in terms of grammar, pronunciation, and vocabulary (Milroy & Milroy, 1999; St. Clair, 1982). Examples of standard varieties include Standard American English in the United States and British Received Pronunciation in the United Kingdom. Nonstandard varieties, in contrast, are those that depart from codified norms in some manner (e.g., pronunciation). Examples of nonstandard varieties include most regional (e.g., American Southern English [ASE]) and ethnic varieties (e.g., African American Vernacular English) in a given society, as well as most foreign accents (e.g., Spanish accent in the United States).
Status attributions are based primarily on perceptions of socioeconomic status (Fiske et al., 2002; Woolard, 1985). Consequently, because standard varieties tend to be associated with dominant socioeconomic groups within a given society, standard speakers tend to be rated higher on status traits than nonstandard speakers (Fuertes, Gottdiener, Martin, Gilbert, & Giles, 2012). Solidarity attributions, on the other hand, tend to reflect in-group loyalty (Ryan, 1983). Language is an important symbol of social identity and use of the in-group style can enhance feelings of solidarity within one’s own linguistic community (Giles, Bourhis, & Taylor, 1977). Accordingly, people tend to attribute more solidarity to members of their own linguistic community, particularly when that community is characterized by high or increasing ethnolinguistic vitality (Ryan, Giles, & Sebastian, 1982). Vitality is a measure of a group’s “strength” in a given society and is determined by the group’s status (e.g., economic power), demographics (e.g., number and distribution), and institutional support (e.g., political and social representation; Giles & Johnson, 1987). In this way, despite being downgraded on the status dimension, nonstandard varieties can possess covert prestige, with users of those forms sometimes—albeit far from always—attributed more solidarity by members of their own linguistic community (e.g., Luhman, 1990).
Although these general patterns have been demonstrated worldwide and cross-culturally, there is necessarily some variability in listeners’ evaluative reactions, with different social groups often expressing different attitudes toward the same variety. Such intergroup variation in language attitudes has been observed with respect to listeners’ ethnicity (e.g., Lambert et al., 1960), age (e.g., Day, 1980; Giles, Harrison, Creber, Smith, & Freeman, 1983), sex (e.g., Lambert, 1967), regional background (e.g., Coupland & Bishop, 2007), and social class (e.g., Lambert, Frankel, & Tucker, 1966), among others. Based on the theoretical model introduced earlier, intergroup variation in language attitudes can arise due to at least two sources (for a discussion, see Dragojevic, 2016). First, different social groups may have different stereotypes toward a given variety, resulting in different evaluative reactions. Second, different social groups may categorize speakers of a given variety differently, with different categorizations activating different stereotypes and, in turn, prompting different evaluative reactions. This latter source of variation—that is, intergroup variation in categorization—is the primary focus of the present study.
Listeners can categorize speakers at varying levels of specificity and with varying degrees of accuracy (Dragojevic, 2016; Ryan, 1983). For example, an ASE speaker can be categorized in terms of a national identity (i.e., American), a regional identity (i.e., a Southerner), or a local identity (i.e., the specific town she comes from; Dragojevic & Giles, 2014). At all levels of specificity, listeners can and do make errors (e.g., Bayard et al., 2001; Clopper & Pisoni, 2004a, 2004b, 2007; Williams, Garrett, & Coupland, 1999), especially when nonnative varieties are involved. For instance, Lindemann (2003) found that American listeners varied widely in their ability to correctly categorize Korean-accented English speakers, frequently miscategorizing them as Chinese, Indian, Japanese, and Latino—indeed, only 8% of participants correctly categorized the speakers as Korean (see also, e.g., McKenzie, 2008, 2015).
The most proximal predictor of categorization accuracy is, arguably, listeners’ familiarity with the target variety—listeners who are more familiar with the target variety are likely to be more accurate in their categorization of speakers who use that variety than listeners who are less familiar (Dragojevic, 2016; McKenzie, 2015; Ryan, 1983). Familiarity can be influenced by a range of interrelated factors, including listeners’ age (Floccia, Butler, Girard, & Goslin, 2009; Girard, Floccia, & Goslin, 2008), regional background and mobility (Clopper & Pisoni, 2004a), and proficiency in the target language (Giles, Bourhis, & Davies, 1979; Giles, Bourhis, Trudgill, & Lewis, 1974; McKenzie, 2008, 2015), as well as the prevailing linguistic landscape in which listeners live (e.g., Dailey, Giles, & Jansma, 2005; Landry & Bourhis, 1997; Shohamy & Gorter, 2009). Language proficiency may be an especially important determinant of categorization accuracy in multilingual settings, where different ethnolinguistic groups are likely to vary in their proficiency in the target language. All else being equal, groups that are more proficient in the target language are likely to be more familiar with and better able to distinguish between different varieties of that language, resulting in higher categorization accuracy rates relative to those who are less proficient.
In turn, variation in categorization accuracy is likely to be consequential—different categories activate different stereotypes and, as a result, are likely to engender different evaluative reactions (see Dragojevic, 2016). Some speakers may benefit from miscategorization—for example, a speaker of a stigmatized variety is miscategorized as belonging to a less stigmatized group—whereas others may be hurt by it— for example, a speaker of a prestige variety is miscategorized as belonging to a less prestigious group. Support for these claims comes from recent research in the language attitudes domain. For example, McKenzie (2008) found that Japanese listeners who correctly categorized a Standard American English speaker (i.e., as “American”) evaluated her more favorably than those who miscategorized her. Relatedly, Yook and Lindemann (2013) found that Korean listeners who were explicitly informed about an African American Vernacular English speaker’s ethnicity (i.e., African American) evaluated her less favorably than those who were not informed. These studies underscore the importance of social categorization in the language attitudes process, demonstrating that different categorizations of the same speaker engender different evaluative reactions. However, past studies examining the effects of categorization on language attitudes have tended to focus on only a single ethnolinguistic group at a given time, demonstrating that intragroup variation in categorization accuracy can produce intragroup variation in language attitudes. To the best of our knowledge, no studies have explicitly examined how different ethnolinguistic groups may vary in their categorization accuracy of a given variety and how that variation may produce systematic, intergroup variation in language attitudes. To address this gap in the literature, the present study investigated intergroup variation in categorization accuracy and its attitudinal consequences.
The Present Study
This study examined language attitudes in Georgia, a small country (population ~3.7 million) in the Caucasus region of Eurasia. Formerly part of the Soviet Union, Georgia gained its independence in 1991. It is a multiethnic and multilingual country. Although ethnic Georgians constitute the majority of the population (86.8%), sizeable minorities of Azerbaijanis (6.3%), and Armenians (4.5%) also reside in the country (GeoStat: National Statistics Office of Georgia, 2016). The country’s sole official language is Georgian. 1 Standard Georgian is based on the Kartluri dialect spoken in the eastern region of Kartli where the capital, Tbilisi, is located (Shosted & Chikovani, 2006). In addition to Georgian, however, substantial portions of the country’s population—particularly minority ethnic groups—also speak other languages, including Armenian, Azerbaijani, and Russian.
Despite its sole official status, proficiency in the Georgian language varies widely among Georgia’s ethnic groups. In general, Georgians tend to be considerably more proficient in the Georgian language than other ethnic groups residing in the country. During the Soviet period, Russian was the de facto lingua franca between most ethnic groups in the region, and there was little incentive for ethnic minorities to learn Georgian. Following independence from the Soviet Union, and during the past decade in particular, the Georgian government undertook efforts to increase proficiency in the state language among ethnic minorities with some success (Berglund, 2016). Nevertheless, Georgian fluency remains low among the Armenian and Azerbaijani populations in Georgia, and these groups primarily speak their respective heritage languages (i.e., Armenian and Azerbaijani) at home and many still receive primary schooling in these languages (Driscoll, Berglund, & Blauvelt, 2016). 2
The present study examined language attitudes among Georgia’s three largest ethnolinguistic groups—Georgians, Armenians, and Azerbaijanis—toward two varieties of spoken Georgian: Tbilisi-accented Georgian and Mingrelian-accented Georgian. Tbilisi-accented Georgian is the standard variety and indexes a Tbiliseli identity—that is, a Georgian from Tbilisi, the country’s capital. Tbiliselis have high ethnolinguistic vitality in Georgia. Approximately one third (~1.1 million) of Georgia’s population resides in the capital, making Tbiliselis the country’s largest regional subgroup. Tbiliselis also tend to be socially and economically advantaged relative to other regional subgroups. Most of Tbilisi’s population (97%) is urban and the region boasts the highest average household income in the country (GeoStat: National Statistics Office of Georgia, 2014). Tbiliselis are also well-represented institutionally, owing, in part, to the fact that most government, media, and other major social institutions are headquartered in the capital.
Mingrelian-accented Georgian is a nonstandard variety and indexes a Mingrelian identity. Mingrelians are a regional and linguistic subgroup of ethnic Georgians who live primarily in the western region of Samegrelo (also known as Mingrelia), situated on the Black Sea coast (see Broers, 2012). Mingrelians have moderate ethnolinguistic vitality in Georgia. Demographically, Mingrelians (population ~450,000) constitute a substantial regional and linguistic subgroup. However, despite their demographic strength, Mingrelians tend to be socially and economically disadvantaged. Most of Samegrelo’s population (60%) is rural, and the region’s average household income is one of the lowest in the country (GeoStat: National Statistics Office of Georgia, 2014). Indeed, the term margali, “A Mingrelian,” has historically been used synonymously with the word “peasant” and to describe someone “who speaks as they speak in the village,” associations that continue to this day (Broers, 2012). Mingrelians also have limited institutional support within Georgia, and any claims that Mingrelians are a separate ethnolinguistic group or that Mingrelian is a separate language (rather than a regional dialect) are vehemently opposed by most Georgians.
Past research suggests that the Mingrelian accent is a stigmatized variety, associated with predominantly negative stereotypes, whereas the Tbilisi accent is a prestige variety, associated with predominantly positive stereotypes. For example, in a recent study, Dragojevic, Berglund, and Blauvelt (2015) found that Tbiliseli and Mingrelian informants both rated Tbilisi-accented speakers more favorably on status and solidarity traits than Mingrelian-accented speakers. Expecting to replicate these results in the present study, we predicted that:
Notwithstanding this general pattern, we also expected the magnitude of this evaluative difference to vary across the three groups, due to variation in the groups’ categorization accuracy of the two accents. As noted earlier, one important factor likely to influence categorization accuracy is listeners’ proficiency in the target language (McKenzie, 2008, 2015; Ryan, 1983). Accordingly, given that the Tbilisi and Mingrelian accents are both varieties of Georgian, listeners’ ability to correctly categorize the two accents is likely to depend, at least in part, on their proficiency in the Georgian language. Given that Georgians tend to be, on average, more proficient in the Georgian language than Armenians and Azerbaijanis, Georgians are likely to be better able to distinguish between and correctly identify both accents than the other two groups.
Based on this rationale, we predicted that:
In turn, we expected this variation in categorization accuracy to be consequential. As noted earlier, different categorizations activate different stereotypes and, in turn, prompt different evaluations (e.g., McKenzie, 2008, 2015; for a discussion, see Dragojevic, 2016). Tbiliselis tend to be positively stereotyped within Georgia and, arguably, more positively than most other regional subgroups. Accordingly, when Tbilisi-accented speakers are miscategorized for another regional subgroup (e.g., Mingrelians), they are likely to be evaluated, on average, less favorably than when they are categorized correctly (i.e., as Tbiliseli). In contrast, Mingrelians tend to be negatively stereotyped within Georgia and, arguably, more negatively than most other regional subgroups. Accordingly, when Mingrelian-accented speakers are miscategorized for another regional subgroup (e.g., Tbiliselis), they are likely to be evaluated, on average, more favorably than when they are categorized correctly (i.e., as Mingrelian). In other words, Mingrelian-accented speakers are likely to (evaluatively) benefit from miscategorization, whereas Tbilisi-accented speakers are likely to be hurt by it. Based on this rationale, we predicted that:
To the extent that different categorizations elicit different evaluations (Hypotheses 4 and 5), then intergroup variation in categorization accuracy (Hypotheses 2 and 3) should produce intergroup variation in language attitudes, qualifying the relationship specified in Hypothesis 1. Specifically, to the extent that (a) Tbilisi-accented speakers are evaluated more favorably when they are categorized correctly than when they are miscategorized and that (b) Georgians are more accurate than Armenians and Azerbaijanis in their categorization of Tbilisi-accented speakers, then it follows that Georgians will, on average (i.e., overall, as a group), evaluate Tbilisi-accented speakers more favorably than Armenians and Azerbaijanis. Conversely, to the extent that (a) Mingrelian-accented speakers are evaluated less favorably when they are categorized correctly than when they are miscategorized and that (b) Georgians are more accurate than Armenians and Azerbaijanis in their categorization of Mingrelian-accented speakers, then it follows that Georgians will, on average (i.e., overall, as a group), evaluate Mingrelian-accented speakers less favorably than Armenians and Azerbaijanis. In other words, although all three groups are likely to evaluate Tbilisi-accented speakers more favorably than Mingrelian-accented speakers (Hypothesis 1), this evaluative difference should be especially pronounced among Georgians. Accordingly, we predicted that:
Method
Participants
Completed questionnaires were obtained from 1,915 Georgian nationals 3 (51.1% women) who ranged in age from 11 to 25 years (M = 16.40). Based on their self-reported ethnicity, participants were classified as belonging to one of three ethnolinguistic groups, who were comparable in terms of age and gender composition: Georgian (n = 780; 54.4% women; Mage = 16.07), Armenian (n = 591; 52.6% women; Mage = 16.77), and Azerbaijani (n = 544; 44.9% women; Mage = 16.44). As expected, the three groups varied in their Georgian language proficiency, F(2, 1,879) = 1125.85, p < .001, η p 2 = .55 (for items used to assess proficiency, see below). Georgians reported being more proficient (M = 5.92) than Armenians (M = 3.64) who, in turn, reported being more proficient than Azerbaijanis (M = 3.46; all ps < .001).
Voice Stimuli
Voice stimuli were produced using the matched-guise technique (Lambert et al., 1960), which involves bidialectical or bilingual speakers producing the same passage of text in different language varieties (e.g., accents), or guises. This procedure ensures that differences across guises reflect only the features of the language variety itself, rather than extraneous variables (e.g., pitch, speech rate). Voice stimuli for the present study were produced by a male speaker in his 20s who was born and raised in Tbilisi, but has a Mingrelian mother and spent summers in Zugdidi (Samegrelo) as a child. He normally speaks with a Tbilisi accent and in his Mingrelian guise has a very strong accent. He was recorded reading a short passage on Euclidian geometry in both Tbilisi-accented Georgian (his habitual speech) and Mingrelian-accented Georgian, yielding two recordings of comparable length. 4
Procedure
Testing was conducted in school classrooms. Participants first provided standard demographic information (e.g., age, sex) and reported their Georgian language proficiency by indicating the extent to which they understand and speak the language (1 = not at all, 7 = completely). Unsurprisingly, participants’ responses to these two latter items were strongly correlated (r = .93, p < .001); consequently, the two items were averaged to form an overall index of Georgian language proficiency (M = 4.52; SD = 1.58). 5
Having completed these preliminary questions, participants listened to the two recordings described above, as well as 11 other “filler” voices—representing other regional and ethnic varieties of Georgian—which were included to avoid potential recognition of the two target guises as belonging to the same speaker (as in, e.g., Giles & Marsh, 1979; Lambert et al., 1960). The voices were presented to participants in a fixed random order in which the third and tenth voice corresponded to the Mingrelian- and Tbilisi-accented guises, respectively. 6 Only results pertaining to these two target voices are reported in this article. To further ensure that participants perceived the voices as belonging to different speakers, each voice was preceded by a different “name tag,” ostensibly representing the speaker’s first and last name; the name tags were conveyed verbally to participants by the experimenter prior to each recording. The Tbilisi-accented guise was preceded by a typical Georgian name, “Shalva Kapanadze,” whereas the Mingrelian-accented guise was preceded by a typical Mingrelian name, “Paata Kvaratskhelia.”
During a brief pause after each recording, participants were asked to rate each speaker on six personality traits adapted from past research (Dragojevic et al., 2015)—three relating to status (intelligent, educated, and cultured) and three relating to solidarity (pleasant, attractive, amusing)—using a 6-point scale (1 = very little, 6 = very much), as well as indicate where they thought each speaker was from via an open-ended question (i.e., Which region or town in Georgia do you think that this person comes from?). Having rated all the voices, participants indicated their ethnicity via an open-ended question.
Given that participants belonged to three different ethnolinguistic groups and varied in their Georgian language proficiency, they were given the option to complete the questionnaire in Georgian, Armenian, Azerbaijani, or Russian. Unsurprisingly, the vast majority of Georgian respondents opted to complete the questionnaire in Georgian (92.7%), followed by Russian (6.4%), Armenian (0.5%), and Azerbaijani (0.4%). Most Armenian respondents opted to complete the questionnaire in Armenian (71.9%), followed by Russian (24.2%) and Georgian (3.9%). Finally, the majority of Azerbaijani respondents opted to complete the questionnaire in Azerbaijani (95.6%), followed by Georgian (3.5%), and Russian (0.9%).
Results
Scale Construction
The six items intended to measure status and solidarity were submitted to a principal components analysis with promax (i.e., oblique) rotation for each guise separately. A single-factor solution emerged in both cases, accounting for 53.70% of the variance for the Tbilisi-accented guise and 56.01% of the variance for the Mingrelian-accented guise. For both guises, all six items loaded highly on this factor (.48-.82). Accordingly, for each guise, the six items were averaged to form an overall evaluation composite score, which was reliable as assessed by Cronbach’s alpha (αTbilisi = .82; αMingrelian = .84). 7
Focal Analyses
Data analysis proceeded in three phases. First, we assessed overall evaluations of the two guises and examined whether they differed across the three groups. To further decompose these evaluations, we then examined if and how the groups differed in their categorization accuracy of the guises. Finally, we assessed if differences in categorization accuracy were consequential—that is, if and how categorization accuracy influenced each group’s evaluations of the guises.
Overall Evaluations
Hypothesis 1 predicted that all three groups would evaluate the Tbilisi-accented guise more favorably than the Mingrelian-accented guise. Hypotheses 6a and 6b predicted that this effect would be especially pronounced among Georgians, who would evaluate the Tbilisi-accented guise more favorably and the Mingrelian-accented guise less favorably than both Armenians and Azerbaijanis. To test these predictions, overall evaluations were submitted to a 2 (accent) × 3 (ethnolinguistic group) analysis of variance, with accent treated as the within-subjects factor and ethnolinguistic group as the between-subjects factor. Significant main effects for accent, F(1, 1,912) = 713.56, p < .001, η p 2 = .27, and ethnolinguistic group, F(2, 1,912) = 24.08, p < .001, η p 2 = .03, were subsumed by a two-way interaction, F(2, 1,912) = 262.16, p < .001, η p 2 = .22. Consistent with Hypothesis 1, all three groups evaluated the Tbilisi-accented guise more favorably (MGeorgians = 4.63; MArmenians = 4.15; MAzerbaijanis = 4.30) than the Mingrelian-accented guise (MGeorgians = 3.05; MAmernians = 3.77; MAzerbaijanis = 4.06; all ps < .001), but that this effect was especially pronounced among Georgians. Specifically, and consistent with Hypotheses 6a and 6b, Georgians evaluated the Tbilisi-accented guise more favorably, but the Mingrelian-accented guise less favorably, than both Armenians and Azerbaijanis (all ps < .001). In addition, Armenians evaluated both guises less favorably than Azerbaijanis (ps < .01). These results are displayed visually in Figure 1.
Effects of ethnolinguistic group on evaluations of the Tbilisi- and Mingrelian-accented guises.
Differences in Categorization
Hypotheses 2 and 3 predicted that Georgians would be more accurate in their categorization of the Tbilisi- and Mingrelian-accented guise, respectively, than both Armenians and Azerbaijanis. To test these predictions, participants’ responses to the open-ended question asking where each guise was from were coded for accuracy. For the Tbilisi-accented guise, participants’ responses were coded as “correct” if they indicated the speaker was from Tbilisi; all other responses, including missing responses, were coded as “incorrect.” For the Mingrelian-accented guise, participants’ responses were coded as “correct” if they indicated the speaker was from Samegrelo (or a specific city in that region); all other responses, including missing responses, were coded as “incorrect.” Each groups’ most frequent categorizations for each guise are displayed in Table 1.
Categorization of the Tbilisi- and Mingrelian-Accented Guise by Georgians, Armenians, and Azerbaijanis.
Note. Values correspond to percentage of participants within each ethnolinguistic group categorizing each guise as belonging to a particular regional group. Bolded percentages correspond to a “correct” categorization for that guise; values in plain typeface correspond to an “incorrect” categorization for that guise.
As expected, categorization accuracy of the Tbilisi-accented guise varied as a function of listeners’ ethnolinguistic group membership, χ2(2) = 111.74, p < .001. Consistent with Hypothesis 2, Georgians were more accurate in their categorization of the guise (56.8%) than both Armenians (35.9%) and Azerbaijanis (29.8%; ps < .05). In addition, Armenians were more accurate than Azerbaijanis (p < .05).
Also as expected, categorization accuracy for the Mingrelian-accented guise varied as a function of listeners’ ethnolinguistic group membership, χ 2 (2) = 976.21, p < .001. Consistent with Hypothesis 3, Georgians were considerably more accurate in their categorization of the guise (79.9%) than both Armenians (13.9%) and Azerbaijanis (5%; ps < .05). In addition, Armenians were more accurate than Azerbaijanis (p < .05).
Effects of Categorization on Evaluations
Hypothesis 4 predicted that the Tbilisi-accented guise would be evaluated more favorably when it was correctly categorized (i.e., as Tbiliseli) than when it was miscategorized. Hypothesis 5 predicted that the Mingrelian-accented guise would be evaluated less favorably when it was correctly categorized (i.e., as Mingrelian) than when it was miscategorized. To test these predictions, evaluations for each guise were submitted to a 2 (categorization accuracy) × 3 (ethnolinguistic group) analysis of variance.
For the Tbilisi-accented guise, significant main effects for categorization accuracy, F(1, 1,909) = 25.83, p < .001, η p 2 = .01, and ethnolinguistic group, F(2, 1,909) = 31.65, p < .001, η p 2 = .03, emerged; the interaction was nonsignificant (p = .25). Consistent with Hypothesis 4, participants who correctly categorized the guise evaluated it more favorably (M = 4.57) than those who miscategorized it (M = 4.26). In addition, post hoc comparisons (two-tailed, Bonferroni adjustment) showed that, regardless of categorization accuracy, Georgians evaluated the guise more favorably (M = 4.63) than Azerbaijanis (M = 4.30) who, in turn, evaluated it more favorably than Armenians (M = 4.15; ps < .05). These results are visually depicted in Figure 2.
Effects of categorization accuracy and ethnolinguistic group on evaluations of the Tbilisi-accented guise.
For the Mingrelian-accented guise, significant main effects for categorization accuracy, F(1, 1,909) = 32.87, p < .001, η p 2 = .02, and ethnolinguistic group, F(2, 1,909) = 19.43, p < .001, η p 2 = .02, were subsumed by a significant two-way interaction, F(2, 1,909) = 3.59, p < .05, η p 2 = .01. Consistent with Hypothesis 5, all three groups evaluated the guise less favorably when they categorized it correctly (MGeorgian = 2.98; MArmenian = 3.07; MAzerbaijani = 3.65) than when they miscategorized it (MGeorgian = 3.34; MArmenian = 3.88; MAzerbaijani = 4.08; all ps < .001). Post hoc comparisons (two-tailed, Bonferroni adjustment) further showed that, among participants who miscategorized the guise, Georgians evaluated it less favorably than Armenians who, in turn, evaluated in less favorably than Azerbaijanis. In contrast, among participants who correctly categorized the guise, Georgians evaluated it less favorably than Azerbaijanis (but not Armenians). These results are visually depicted in Figure 3.
Effects of categorization accuracy and ethnolinguistic group on evaluations of the Mingrelian-accented guise.
Discussion
The present study examined the role of social categorization in the language attitudes process. Participants from three different ethnolinguistic groups living in Georgia that varied in their Georgian language proficiency—Georgians, Armenians, and Azerbaijanis—listened to a speaker reading a short passage of text in both a Tbilisi-accented Georgian (standard variety) guise and Mingrelian-accented Georgian (nonstandard variety) guise. We predicted that the three groups would vary in their ability to correctly categorize the two guises and that this intergroup variation in categorization accuracy would be consequential and result in intergroup variation in language attitudes. These hypotheses were supported.
The three groups varied in their categorization accuracy of the two guises. Georgians, who were most proficient in the Georgian language, were more accurate in their categorization of both guises than Armenians and Azerbaijanis, both of whom were less proficient in the Georgian language. With respect to the Tbilisi-accented guise, 56.8% of Georgians categorized it correctly, compared with 35.9% of Armenians and 29.8% of Azerbaijanis. Group differences in categorization accuracy were even more pronounced for the Mingrelian-accented guise—whereas the vast majority (79.9%) of Georgians correctly categorized the guise as “Mingrelian,” only 13.9% of Armenians and 5% of Azerbaijanis did. Patterns of miscategorization also varied across the three groups (see Table 1). For example, whereas Armenians (29.1%) and Azerbaijanis (31.1%) frequently miscategorized the Mingrelian-accented guise as “Tbiliseli,” Georgians almost never did (2.9%).
Variation in categorization accuracy was consequential: Listeners evaluated the guises differently depending on whether they categorized them correctly or incorrectly. Specifically, all three groups evaluated the Tbilisi-accented guise more favorably when they categorized it correctly (i.e., as Tbiliseli) than when they miscategorized it, suggesting they have more favorable stereotypes toward Tbiliselis than other regional subgroups. Conversely, all three groups evaluated the Mingrelian-accented guise less favorably when they categorized it correctly (i.e., as Mingrelian) than when they miscategorized it, suggesting they have less favorable stereotypes toward Mingrelians than other regional subgroups.
Intergroup variation in categorization accuracy, in turn, resulted in intergroup variation in language attitudes. Although all three groups evaluated the Tbilisi-accented guise more favorably than the Mingrelian-accented guise, this evaluative difference was especially pronounced among Georgians. Specifically, Georgians evaluated the Tbilisi-accented guise more favorably, and the Mingrelian-accented guise less favorably, than both Armenians and Azerbaijanis, due in part to Georgians’ higher categorization accuracy of both guises. These findings have theoretical, methodological, and practical implications.
Theoretical Implications
As noted at the outset, language attitudes have been theorized to reflect two sequential cognitive processes: categorization and stereotyping (Lambert et al., 1960; Ryan, 1983). That is, listeners use language cues (e.g., accent) to make inferences about speakers’ social group membership(s) and, in turn, attribute to them stereotypic traits associated with those inferred group memberships. The present study lends strong support to this theoretical model—and particularly the mediating role of social categorization in the language attitudes process—by demonstrating that different categorizations of the same speaker engender different evaluative reactions (see also, Dragojevic & Giles, 2014). Moreover, and relatedly, results of the present study provide no evidence to support the argument that language attitudes reflect intrinsic differences between language varieties, such as their supposed linguistic superiority or aesthetic quality (for a discussion and critique of the “inherent value hypothesis,” see Giles & Niedzielski, 1998). Indeed, if language attitudes were simply a response to intrinsic differences between language varieties, then social categorization would be immaterial to the language attitudes process—results of the present study, along with other research (e.g., Giles et al., 1974; McKenzie, 2008; Yook & Lindemann, 2013), clearly show that is not the case.
Furthermore, results of the present study also contribute to our theoretical understanding of when and why intergroup variation in language attitudes may emerge. As noted earlier, past research has found that different social groups often express different attitudes toward the same language variety (see Giles & Watson, 2013). The present study suggests that, at least in some situations, such intergroup variation in language attitudes may stem directly from intergroup variation in categorization. That is, different social groups may evaluate the same speaker differently because they categorize the speaker differently. The most proximal predictor of categorization accuracy is likely to be listeners’ familiarity with the target variety which, in turn, can be influenced by a range of interrelated factors, including listeners’ age, regional background, and language proficiency (for a discussion, see Dragojevic, 2016; Ryan, 1983).
Notwithstanding the above, it is important to acknowledge that the intergroup variation in language attitudes observed in the present study was not entirely explained by intergroup variation in categorization. Indeed, regardless of categorization accuracy, the three groups differed in how they evaluated the two accents, suggesting that they also have different stereotypes toward the varieties in question. For example, among participants who correctly categorized the Tbilisi-accented guise, Georgians evaluated it more favorably than both Armenians and Azerbaijanis, suggesting that Georgians have more favorable stereotypes toward Tbiliselis than do the other two groups. Similarly, among participants who correctly categorized the Mingrelian-accented guise, Georgians evaluated it less favorably than Azerbaijanis (but not Armenians), suggesting that Georgians have less favorable stereotypes toward Mingrelians than do Azerbaijanis. These findings are also fully consistent with the theoretical model outlined above, demonstrating that intergroup variation in language attitudes can arise due to both (a) intergroup variation in categorization and (b) intergroup variation in stereotypes.
Methodological Implications
Results of the present study also have methodological implications. As noted earlier, the dominant paradigm in language attitudes research requires respondents to listen to audio-recorded voices of a number of speakers and rate each on various traits. Past studies operating within this “speaker evaluation” paradigm, however, have rarely asked respondents to identify which social group(s) they think the speakers they are rating belong to and instead have assumed that respondents are categorizing the speakers as intended (i.e., correctly). Consistent with previous research (e.g., Lindemann, 2003; McKenzie, 2008, 2015), results of the present study suggest that, at least in some situations, this may not be a tenable assumption and that important individual and group differences in categorization may emerge (for a discussion, see Preston, 1999; Ryan, 1983). To the extent that variation in categorization is consequential—as demonstrated by the present study—researchers working in this domain should be more cognizant of the role that categorization plays in the language attitudes process and measure it explicitly (for similar calls, see, e.g., Preston, 1999).
We are not suggesting that studies that do not explicitly assess categorization are any less valid than those that do. Rather, we are suggesting that the former may, in some situations, lend themselves to interpretive difficulties. For example, suppose that American listeners rate a speaker with accent “X” favorably. Based on this, we could conclude that Americans rate this accent in this particular way, but we could not conclude, however, that Americans have positive stereotypes toward group “X” because the majority of listeners may not have categorized the speaker as belonging to group “X.” Suppose further that Australian listeners rate the same speaker unfavorably. Although we could conclude that Australians rate the accent less favorably than Americans, we could not conclude, however, that Australians have negative stereotypes toward group “X” or even that they have more negative stereotypes toward group “X” than Americans because, once again, the majority of listeners may not have categorized the speaker as belonging to group “X.”
Measuring categorization explicitly not only aids in the interpretation of findings but also makes it possible to more directly compare findings across studies. Items assessing categorization can easily be incorporated into existing language attitudes questionnaires. Categorization can be measured using an open-ended format in which listeners are asked to write down which social group(s) they think the speakers they are rating belong to (see, e.g., Lindemann, 2003; McKenzie, 2008), or a closed-ended format in which listeners are presented with a list of possible group memberships and asked to mark the one(s) they deem appropriate (see, e.g., Bayard et al., 2001). Each method has advantages and disadvantages. The closed-ended format reduces the number of possible responses, making comparisons within and across studies relatively straightforward; at the same time, it imposes researchers’ own assumptions about possible categorizations, which may or may not accurately reflect listeners’ own perceptions. The open-ended format has the advantage of more faithfully capturing listeners’ own perceptions, which are likely to vary both in accuracy and specificity due to a host of factors; at the same time, it is likely to produce more idiosyncratic responses that may be difficult to compare within and across studies. Despite these shortcomings, and given that categorization is a subjective process that is inherently prone to variability, we favor the open-ended format.
Practical Implications
Our results also have practical implications. Namely, to the extent that language attitudes are a product of social categorization processes—which is inherently prone to variation—results of the present study suggest that, outside the laboratory, language attitudes may be far more variable and dynamic than existing research suggests (see also, Dragojevic & Giles, 2014). That is, the same speaker may garner markedly different evaluations in different situations, simply because they may be categorized differently. How listeners categorize a given speaker can be influenced not only by listeners’ familiarity with the target variety (as noted above) but also by other factors, including the presence of nonlinguistic cues to social group membership(s) and the social comparative context the speaker is encountered in, among others.
Although language often emerges as one of the most potent cues to social categorization (Kinzler, Shutts, DeJesus, & Spelke, 2009; Rakić, Steffens, & Mummendey, 2011; for a discussion, see Kinzler, Shutts, & Correll, 2010), it is certainly not the only one. In many everyday situations, listeners may have a range of other, nonlinguistic cues, which they can base their guesses about speakers’ social group membership(s) on, including speakers’ skin color, dress, and so forth. These nonlinguistic cues may interact with or, in some cases, override linguistic cues as the basis for social categorization, resulting in markedly different evaluative reactions (e.g., Giles & Sassoon, 1983; Lambert, 1967; Rubin, 1992; Yook & Lindemann, 2013).
The context in which evaluations occur can also influence how speakers are categorized and, thus, evaluated. Social categorization is a subjective process that depends on the contrasts that are perceptually most obvious (i.e., accessibility) and meaningful (i.e., fit) in a given social comparative context (see self-categorization theory: Turner, Hogg, Oakes, Reicher, & Wetherell, 1987). As the nature and distribution of social stimuli changes, so does the relative accessibility and fit of different social categories. In this vein, Dragojevic and Giles (2014) found that Californian listeners evaluated ASE speakers more favorably when the speakers were contrasted with a Punjabi-accented than a Californian-accented speaker, presumably because the first situation made a national identity salient in which ASE speakers constituted the in-group (i.e., Americans), whereas the second made a regional identity salient in which ASE speakers constituted the out-group (i.e., Southerners; see also Abrams & Hogg, 1987).
Limitations
The present study has several limitations. First, it examined attitudes toward only two language varieties among three different ethnolinguistic groups. Future research should examine how other groups vary in their ability to correctly categorize different language varieties and what effects (if any) that variation has on listeners’ language attitudes. Second, and related, results of the present study are based on listeners’ ratings of voice stimuli produced by only a single speaker. In a previous study, Dragojevic at al. (2015) used these same voice stimuli in conjunction with corresponding Tbilisi- and Mingrelian-accented voice stimuli produced by two other speakers. They found that Tbiliseli and Mingrelian respondents rated all three speakers (including the one from the present study) more favorably in their Tbilisi-accented guise than their corresponding Mingrelian-accented guise, suggesting that the findings pertaining to overall evaluations of the guises reported herein are generalizable to other speakers. However, that study did not assess listeners’ categorization accuracy of the guises. Accordingly, future research should attempt to extend the findings obtained in the present study to other speakers as well.
Third, in an effort to ensure that listeners perceived the two voices of interest as belonging to different speakers, each voice was preceded by a corresponding ethnic “name tag,” ostensibly representing the speaker’s first and last name. We acknowledge that these name tags may have functioned as an additional cue to speakers’ social group membership(s) and may have influenced how listeners categorized the speakers. Accordingly, we cannot unequivocally attribute categorization accuracy solely to accent, as it may have been based on both cues (i.e., accent and name tags). Indeed, had the voices not been preceded by corresponding name tags, we contend that categorization accuracy rates would have been even lower than what we observed, particularly among Armenian and Azerbaijani respondents who are less proficient in the Georgian language than ethnic Georgians. As such, our method represents a “hard test” of sorts, which shows that listeners are not necessarily accurate in their categorization of speakers even when redundant cues to speakers’ identity are made available to them (as they would be in many real-world situations). Future research should examine how multiple cues to speakers’ social group membership(s)—which may complement or contradict one another—influence the categorization process. Nonetheless, we emphasize that this has no bearing on the validity or interpretation of the key findings of the present study—that is, different ethnolinguistic groups varied in their categorization accuracy of the two guises, and intergroup variation in categorization accuracy resulted in intergroup variation in language attitudes. In other words, regardless of whether listeners based their categorization on the speaker’s accent, ostensible name, or both, the fact remains that they varied in accuracy and that different categorizations of the same guise engendered different evaluations, lending strong support to the argument that social categorization mediates the language attitudes process (Lambert, 1967; Ryan, 1983).
Conclusion
Results of the present study highlight the central role of social categorization in the language attitudes process and lend themselves to three general conclusions. First, different social groups can vary in their ability to correctly categorize speakers based on speakers’ language. One factor influencing groups’ categorization accuracy is their proficiency in the target language: Groups who are more proficient in a given language are better able to differentiate between and identify varieties of that language, relative to those who are less proficient. Second, variation in social categorization is consequential: Different categorizations activate different serotypes and, consequently, engender different evaluative reactions. Third, to the extent that groups vary in their categorization accuracy and that different categorizations activate different stereotypes, then intergroup variation in categorization accuracy can result in intergroup variation in language attitudes. In other words, one reason different social groups may evaluate the same language variety differently is because they categorize that variety differently. To the best of our knowledge, this is the first study to empirically demonstrate that intergroup variation in categorization accuracy can result in intergroup variation in language attitudes.
Footnotes
Acknowledgements
We would like to thank two anonymous reviewers and Howie Giles for their insightful comments, which improved this article.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Field research was supported by grants from Sixten Gemzéus stiftelse, the Swedish Foundation for International Cooperation in Research and Higher Education (STINT), Forskraftstiftelsen Theodor Adelswärds Minne, Byzantiska resestipendiet, and Borbos Hanssons stipendiefond.