Authors: Nitzan Trainin (Department of Linguistics, Tel Aviv University; Sagol School of Neuroscience, Tel Aviv University), Einat Shetreet (Department of Linguistics, Tel Aviv University; Sagol School of Neuroscience, Tel Aviv University)
Categories: Regular Article, Alignment, Social meaning, Speaker‐specific, Learning, Generalization, Stereotypes
Source: Cognitive Science
Doi: 10.1111/cogs.70086
We Do Not Speak Like This Here: The Role of Perceived Foreignness in Shaping Speaker‐Specific Social and Linguistic Inferences
Authors: Nitzan Trainin, Einat Shetreet
Abstract
People use many kinds of cues that help them navigate social interactions. We examined how perceived foreignness affected people's ability to map speaker‐specific naming preferences, align with their interlocutors concerning these preferences, and make social inferences based on them. In a pseudo‐interactive experiment, participants engaged with two simulated one with a common native name who consistently used favored words, and one who consistently used the disfavored alternatives, and had either a native name, a foreign name associated with positive stereotypes (American), or a foreign name associated with negative stereotypes (Former Soviet Union; FSU). We assessed participants’ tendencies to align with each speaker's lexical choices, their ability to generalize disfavored lexical use to other sorts of language use, and the social inferences they drew about each speaker. Results showed that perceived foreignness modulated both linguistic alignment and social judgments. The alignment effect was larger for FSU and native speakers compared to the American speakers. Interestingly, this stemmed from the increased tendency to use the disfavored words with the common native speaker when the uncommon speaker was American, suggesting that speakers’ nationality modulated words’ perceived disfavoredness. Further, generalizations about social traits (e.g., cooperativeness) varied by nationality, with American speakers rated more positively despite similar linguistic behaviors. These findings reveal that foreignness‐associated stereotypes can modulate the social consequences of language use, suggesting a bidirectional dynamics where social identity both shapes language processing and is shaped by it. This extends theories of social meaning by demonstrating how social expectations conditionally interact with linguistic behaviors.
Introduction
It is almost trivial to say that not all speakers speak the same. If two different people walk into a store, one of them might say she wants to buy an item, and the other one might say she wants to purchase it. In the context of selling that item, it should probably not matter to the vendor who used each word, but learning to associate each speaker with their word choice might assist in many other contexts to better understand speakers during interaction and to predict their behavior. The current study concerns speaker specificity in the context of interacting with non‐native speakers.
Speaker‐specificity
Speaker‐specificity has been the focus of many studies that show successful mapping of a person with their language use (Creel, Aslin, & Tanenhaus, 2008; Eisner & Mcqueen, 2005; Grodner & Sedivy, 2011; Kamide, 2012; Kraljic & Samuel, 2007; Kroczek & Gunter, 2017; McLennan & Luce, 2005; Pogue, Kurumada, & Tanenhaus, 2016; Schuster & Degen, 2020; Trainin & Shetreet, 2024, 2025). Mapping interspeaker variability in language use could help in understanding people's linguistic and nonlinguistic behavior (Trainin & Shetreet, 2025), both within‐context and out‐of‐context (i.e., outside the specific interaction).
One of the most prevalent findings related to speaker‐specificity in the literature concerns people's tendency to align with their partners’ referential choices (e.g., Brennan & Clark, 1996; Brown‐Schmidt, 2009; Clark & Wilkes‐Gibbs, 1986; Garrod & Anderson, 1987; Metzing & Brennan, 2003). In testing this well‐studied phenomenon, some studies used amorphic, unnamed stimuli (i.e., tangrams; e.g., H. H. Clark & Wilkes‐Gibbs, 1986), and others used real‐world objects that could be described using a favored word or a disfavored synonym (Garrod & Anderson, 1987; Tobar‐Henríquez et al., 2020; Tobar‐Henríquez, Rabagliati, & Branigan, 2021). In the latter case, the favored words were used by one speaker and the disfavored ones, by a different speaker (e.g., “pistol” vs. “gun” in British English; Branigan, Pickering, Pearson, McLean, & Brown, 2016, 2011; Hopkins, Yuill, & Branigan, 2017; Tobar‐Henríquez et al., 2020; Tobar‐Henríquez et al., 2021). In these studies, participants were more likely to use the disfavored words when interacting with the speaker who used them themselves (but not more likely than using the favored words). That is, although these words are not typically preferred by most individuals, people still tend to align (to some extent) with people who seem to prefer them. This pattern of behavior, which can also be referred to as audience design (H. H. Clark & Murphy, 1982), is also consistent with communication accommodation theory (Dragojevic, Gasiorek, & Giles, 2016; Giles, Taylor, & Bourhis, 1973), according to which individuals adjust their behavior to accommodate their conversational partners’ communicational needs.
Speaker‐specific information is not limited to in‐context usage, as in the case of alignment. It can also inform predictions regarding future behavior—linguistic or social. Interlocutors noticing certain linguistic preferences may expect them to reflect stable tendencies, such as linguistic style or even social stance, and use this to form expectations about the speaker's behavior, even in novel contexts. Drawing conclusions from a certain linguistic use to other characteristics of the speaker could, in turn, contribute to gaining a deeper understanding of them as individuals. Indeed, a recent study showed that people generalized disfavored word choices to other—previously unmentioned—words (Trainin & Shetreet, 2025). This generalization could help people better predict their interlocutor's general linguistic choices and, therefore, would facilitate processing and comprehension. Further, language serves as a social cue (see Social Meaning section below). In the context of word choices, it was found that people inferred individuals’ social characteristics based on such information, for example, a person who used disfavored words was considered less friendly and less cooperative (Trainin & Shetreet, 2025). This could help in forming more accurate predictions about individuals’ social behavior. These social inferences at the individual level have also been demonstrated in studies testing phonetic interspeaker variation (Campbell‐Kibler, 2007; Podesva, Reynolds, Callier, & Baptiste, 2015), Moreover, social inferences from a person's language use at the group level could be achieved by associating a speaker's language use with a certain community or social stereotype (Beltrama & Schwarz, 2022; Clopper & Pisoni, 2007; Labov, 1973; Palomares, Soliz, Gallois, & Giles, 2016).
Social meaning
Social meaning refers to the notion that language use serves to index social features, or in other words—how social information is conveyed by certain types of language use (Bucholtz, 2010; Bucholtz & Hall, 2005; Eckert, 1989, 2008, 2012; Labov, 1963). For example, in Labov's (1963) study of Martha's Vineyard, residents used localized vowel pronunciations to convey social identity and solidarity with the island's traditional community, distancing themselves from the influence of mainland culture and tourism. In recent years, it has become clearer that how language and social meaning interact is bidirectional. That is, social information can be used to predict a certain form of language use, or to infer different meanings based on who the speaker is (Beltrama & Schwarz, 2022; Fairchild & Papafragou, 2018; Mahler, 2020), just as it can be inferred by language use. Individuals in a certain society are usually familiar with which linguistic variants represent each community (e.g., Tobar‐Henríquez et al., 2021). Therefore, we expect interlocutors to be able to (1) use linguistic cues to deduce a person's social background (i.e., using language use to establish social meaning), and (2) predict linguistic forms based on the person's known/assumed social characteristics (i.e., use social cues to establish language use). In this study, we examine both directions.
Using linguistic cues to infer social meaning
The notion that certain linguistic forms are associated with certain social groups is hardly new. From the early 1960s, it has been consistently shown that change in sound features is closely related with belonging to a certain community or social status (Campbell‐Kibler, 2007; Eckert, 2008, 2012; Labov, 1963, 1966, 1972; Trudgill, 1974; Wolfram, 1969). Furthermore, it has been observed that people use varying linguistic forms to index social status (Eckert, 1989; Silverstein, 2003), and that variation in linguistic style reflects both societal and personal changes (Eckert, 2017; Sankoff, 2006). That is, interlocutors can use language to infer many social features about a certain speaker, even without knowing anything about them, including where they live, whether they belong to a lower or upper class, and so on.
Using social features to infer linguistic meaning
The relationship between linguistic variation and social identity also means that when encountering a person, knowing some social features about them might activate linguistic expectations toward them. People constantly use contextual cues to adjust their expectations toward linguistic input (Brothers, Swaab, & Traxler, 2015; Glucksberg, Kreuz, & Rho, 1986; Heller, Grodner, & Tanenhaus, 2008; Mazzarella, Trouche, Mercier, & Noveck, 2018). A large piece of the contextual puzzle can be people themselves and specifically, their social background (Brown‐Schmidt, Yoon, & Ryskin, 2015). Indeed, people use their knowledge regarding a person's political views (Mahler, 2022; Weissman, submitted) or personality (Beltrama & Schwarz, 2022) to adjust the way they infer meaning from this person's utterances. For example, Beltrama and Schwartz (2022) observed that quantities produced by people who were perceived as more “nerdy” were expected to be more precise than those produced by “laid‐back” individuals.
One of the most‐studied social cues that people use to adjust meaning inference is foreignness (Fairchild & Papafragou, 2018; E. Gibson et al., 2017; Hanulíková, van Alphen, van Goch, & Weber, 2012; Lev‐Ari & Keysar, 2010, 2012). For example, Fairchild and Papafragou (2018) showed that people tend to be more forgiving toward infelicitous underinformative utterances when those come from a non‐native speaker. Furthermore, perceived foreignness can have varying types of effects on linguistic interaction. For example, foreign speakers’ utterances are seen as less credible (Lev‐Ari & Keysar, 2010). Moreover, it has been suggested that when conversing with a non‐native speaker, people maintain less‐detailed representations of their linguistic input (Lev‐Ari & Keysar, 2012).
Current study
Recently, Trainin and Shetreet (2025) showed that the usage of a certain set of disfavored words, in addition to modulating the alignment effect, is generalized both to other disfavored words and to social features such as friendliness and cooperativeness. In that study, participants had no a priori information about the speakers, so the effects stemmed directly and only from the speakers’ language use. In the current study, we ask if these effects could be modulated by a priori social information about the speaker. Additionally, we ask whether different attitudes toward different groups of foreign speakers might lead to varying effects on alignment and generalization. An interesting distinction between groups of foreign speakers is that of the Former Soviet Union (FSU) and American immigrants in Israel. FSU immigrants form a large group that has often suffered from prejudice and racism, while Americans are usually linked to positive stereotypes.
FSU‐originated and American people in Israel
During the 1990s, around 800,000 people from the FSU immigrated to Israel (Remennick, 2003), in what was termed “The Great Aliya (homecoming).” This substantial immigration formed an empowered social group, and had a substantial impact on Israeli society (Remennick, 2014). Moreover, Russian‐speaking immigrants were seen as somewhat separatist, partly because of a strong tendency within that group for language maintenance (i.e., keep speaking Russian, though in a Hebrew‐speaking society; Spolsky & Shohamy, 1999). This perceived separatism possibly led Israelis to portray FSU‐originated immigrants and their descendants as associated with negative stereotypes (Remennick, 2014).
Americans, on the other hand, are perceived more positively as a foreign group by Israelis. In recent decades, American Jews have represented a small share of the total immigrants to Israel (less than 5% between 1984 and 2022; Israel's Central Bureau of Statistics—cbs.gov.il, 2023). Similarly to Russian‐speaking immigrants to Israel, and despite their relatively small number, American immigrants have had a significant impact on Israeli society (Rebhun & Waxman, 2000). As the Israeli culture consistently undergoes an Americanization process, American people are associated with positive stereotypes (Avraham & First, 2003; Rebhun & Ari, 2010; Rebhun & Waxman, 2000; Sarna, 2021). Perhaps relatedly, it has been observed that people who learn a second language tend to develop positive stereotypes toward people for whom that language is their first language (El‐Dash & Busnardo, 2001). Because learning English is compulsory in Israel from the fourth grade until high school graduation (Aronin & Yelenevskaya, 2022), it is an additional plausible source of positive stereotypes toward Americans.
All in all, people from the FSU and people from USA substantially differ in the way they are looked upon by the Jewish‐Israeli majority in Israeli society. These differing stereotypes could be reflected both in the way Israeli individuals expect them to use language, and in how their language use could contribute to how they are perceived socially.
In the current study, we aimed to examine how the perceived foreignness of different groups affects (1) the way that people learn to map interspeaker variability in word choices; (2) people's tendency to lexically align with their partners; (3) the generalizations they make regarding a speaker based on their language use. We exposed participants to speakers from different social groups, each with different lexical preferences, and then tested them on the information they stored based on that exposure. Overall, our goal was to investigate the bidirectional link between social identity and language use. That is, we made an attempt to understand how a speaker's social identity shapes the way their language use is perceived and acted upon, and how a speaker's language use impacts their perceived social identity.
Methods
Participants
Ninety‐six native Hebrew speakers (18−39 years old, M = 25.62, SD = 4.14; 68.75% females) were recruited via social media. The inclusion criteria, in addition to speaking Hebrew as a first language, were having been born in Israel and having no reported cognitive or language impairments. Participants were randomly assigned to one of three experimental between‐subject native, American, or FSU (based on the uncommon speaker, more details in Section 2.2). ^1^ The desired sample size (32 for each group) was calculated based on the estimated learning effect (based on the reported effect size in Trainin & Shetreet, 2024), and was carried out using the G^*^Power program (Faul, Erdfelder, Lang, & Buchner, 2007). All participants gave informed consent and were compensated for their participation. This study was approved by Tel Aviv University's ethics committee (approval number 0005280–1).
Stimuli
The experimental stimuli, taken from Trainin and Shetreet (2025), consisted of image‐word pairs in Hebrew, such that each critical image was associated with two alternative Hebrew words—one per experimental confederate (simulated by a computer). Each pair of words consisted of a favored alternative and a disfavored one (at a ratio of ≥ 20), such that one of the confederates (“the uncommon speaker”) consistently produced the disfavored words and the other (“common speaker”) produced their favored alternatives. The ratio was estimated based on a pretest conducted in native speakers (for Trainin & Shetreet, 2025), where participants rated, on a visual analog scale, the probability of using each alternative for a given image. Words considered disfavored when rated as likely to be selected by native speakers for a given image in 20% (or fewer) of the responses. All the image‐word associations were also pretested to make sure that both alternatives could potentially describe the image. Images were identical to the ones used in Trainin and Shetreet (2025) and included real‐world stimuli. Words were presented in a written form, so it would seem as if they were typed by the confederates. Using written language allowed us to eliminate possible effects of comprehensibility (Munro & Derwing, 1995). There were six pairs of critical words, with each word repeating four times per speaker. The items were taken from one of the experimental conditions in Trainin and Shetreet (2025), in which the favorability status of these words was pretested to ensure they were compatible with the images they were paired with. We also included six fillers, each repeating four times, where the same word was used to describe the image by both speakers. The number of repetitions followed Trainin and Shetreet (2025), to replicate the experimental settings as closely as possible. The experiment included a between‐subject manipulation of the speakers’ perceived nationality, varying the identity of the uncommon speaker (only). In the IL condition (the control condition), both simulated speakers had conventional Israeli names (Adi for the female and Yoni for the male); in the other two conditions, the common speaker always had a conventional Israeli name, and the uncommon speaker had a foreign name. In the FSU condition, the foreign name was a Russian one (Katya for the female and Dima for the male), and in the US condition, an American one (Amy for the female and Josh for the male). Thus, the common speaker was always supposed to be perceived as a native Hebrew, Israeli speaker, and the perceived nationality of the uncommon speaker varied between three perceived nationality conditions. Across all perceived nationalities, one speaker was female and the other was male, counterbalanced between common and uncommon speakers. The order of the speakers was counterbalanced across participants.
Procedure
The same procedure from Trainin and Shetreet (2025) was used. To create a sense of interactivity, participants were invited to join a Zoom session, in which they were led to believe that they would cooperate with other naïve participants who joined the same session. They were further instructed that they were assigned to separate “breakout rooms,” so the only interaction would occur in the experimental platform. In the experimental platform, participants were initially introduced to the other participants. In fact, these “participants” were simulated by a computer program. Each participant “met” two simulated participants, one after the other. To ensure that—in all conditions—participants were aware that there were two different speakers, one of the “partners” was male and the other was female. Each participant chose their own avatar (in the excuse of keeping their privacy) and typed a name that would represent them (they were encouraged to use their real names, so the manipulation would be more credible). The simulated confederates were also given an avatar each (one of a female and one of a male), which were identical across participants, but had different names based on the experimental condition (see above).
The experiment was conducted in five steps, presented in a fixed sequence (see Fig. 1). The first two steps, Exposure and Alignment, involved an interactive picture selection task. This task included two Directors and Matchers, such that directors instructed matchers on which image to select (see Fig. 1). Participants always acted as matchers in the first step, and as directors in the second step. After this task, participants answered post‐task questions, which included three steps testing how well the interspeaker variation was mapped, as well as how the interspeaker variation was generalized both linguistically and socially.
Step Exposure. In this step, the participants acted as matchers in an interactive picture‐selection task. In each block, they were instructed by one simulated confederate acting as the director which image to choose. Participants were told that the directors saw only one image at a time, at the center of their screen. They were required to click on the picture that their director described. After each choice, there was feedback (either a green “V” sign if they selected the correct image, or a red “X” sign if they did not). Participants were instructed before the task that after each block, they would receive a score, together with their partner, based on their success in the task. There were two blocks, such that each participant was exposed to two different speakers. To make the speakers even more evidently distinct, in addition to using different genders, the speakers’ utterances were color coded such that words typed by the male speaker were always shown in blue, and words typed by the female speaker were shown in red. There were six critical words and six fillers, with each word repeated four times. At the end of each block, participants received a joint score with their partners, which was the number of correct choices.
Step Alignment. In this step, participants acted as directors in the same interactive picture selection task. They were presented with a single picture at the center of their screen and were asked to instruct—by typing their response—the simulated confederates, now acting as matchers, which image to choose. For participants identified as males, the responses appeared in blue, and for participants identified as females—in red. For each participant, the simulated matchers were the same ones acting as their directors in Step 1. Like in Step 1, there were two blocks, one per speaker. The same order of presentation of speakers as in Step 1 was used (e.g., if the uncommon speaker was the first director in Step 1, s/he was also the first matcher in Step 2). Each image was repeated four times, such that participants had to describe a total of 48 trials (24 critical trials and 24 filler trials). At the end of each round of this task, participants received a joint score with their partners, which was a random number between 46 and 48 (out of 48 trials).
Step Detection test. In this step, on each trial, participants were shown an image at the center of the screen and were asked, about each simulated confederate, whether they had used a certain word to describe a certain image (i.e., “Did X describe the following image using the word Y?”; Yes/No). The order of trials was fully randomized.
Step Linguistic generalization. This step involved a task similar to that of the detection test, with a small tweak. In this step, we presented participants with images that were not included in the picture‐selection task and asked whether it was likely—hypothetically—that a certain speaker would have used a certain utterance (i.e., “Would it be likely for X to describe the following image using the utterance Y?”; Yes/No). The utterances were of one of three types of linguistic phenomena at different linguistic (1) Word‐level: favored/disfavored words in Hebrew, not included in the exposure phase; (2) Phrase‐level: multi‐adjectival noun‐phrases with a favored/disfavored adjective order (Trainin & Shetreet, 2021); and (3) Sentence‐level: sentences with a noncanonical constituent order (VSO; as opposed to the canonical SVO in Hebrew). For each phenomenon, there were two alternative descriptions—one favored and one disfavored. These phenomena were included to evaluate whether linguistic generalizations were made, and possibly even extend beyond the (lexical) level upon which they were formed. The inclusion of these specific structures across different linguistic levels was motivated by their unexpected (or disfavored) linguistic patterns, allowing us to test whether individuals expect such noncanonical forms, beyond the lexical level, from the uncommon speaker (also see Trainin & Shetreet, 2024, 2025). The order of trials was fully randomized for each participant, across all types of utterances.
Step Social generalization. In the last step, we asked about some perceived social and personality traits of each speaker separately. We used a rating task with visual analog scales, one for each speaker, for four questions, as listed below. The questions were either related to the perceived linguistic competence of the speakers (Q1 and Q2), or to their perceived sociability (Q3 and Q4). The two speakers were presented simultaneously, and the order of the questions was fully randomized. Q1)How likely it is that the speaker was born abroad? (ranging from not at all to very)Q2)How many books would you estimate the speaker has read? (ranging from a few to many)Q3)How many friends would you estimate the speaker has? (ranging from a few to many)Q4)How cooperative was the speaker's behavior? (ranging from not at all to very)
Norming studies
Linguistic expectations norming study
A priori linguistic expectations may vary between perceived nationalities, given the foreign language status in two of our conditions. We, therefore, administered a separate norming study (N = 30, 60% females) assessing the a priori linguistic expectations toward the different speakers. For this aim, we conducted a test similar to that in Step 4, but without exposure to language use. That is, on each trial, participants saw a speaker's Avatar, an image, and an utterance that could describe it (a word/a phrase/a sentence; see Section 2.3), and were required to respond “no” if they did not expect the speaker to use it, and “yes” if they did. Each participant saw characters from a single “nationality,” such that 10 participants saw characters with conventional Israeli names (“Adi” and “Yoni”), 10 participants saw ones with Russian names (“Katya” and “Dima”), and 10 saw ones with American names (“Amy” and “Josh”).
Social norming study
Because different populations, as indicated by the different names of the simulated confederate, probably carry some a priori social biases, a comparison is needed between neutral social expectations that arise from encountering characters with foreign or local names and those that arise after exposure to their language use. This second norming study (on a different cohort of participants; N = 30, native Hebrew speakers with no cognitive impairment; 60% females) was similar to the linguistic expectations norming study, but instead of asking about the speakers’ language use, we asked participants to rate, on visual analog scales as a standalone task, the social features of different characters with the same names as those used in main experiment, like in Step 5. Specifically, we asked participants (1) how many books they estimate each character has read; (2) how likely it is for each character to have been born abroad; and (3) how many friends they estimate each character has. We did not include the cooperativeness question in this norming study because asking about the cooperative behavior of a character with whom participants did not have any sort of interaction would be infelicitous.
Statistical analysis
Below, we report the results of each step's analysis separately (except Step 1, which is the exposure phase, where no critical responses were collected). For each step, we lay out the statistical analyses and report the results. In the detection‐test step, we include only perceived nationality as a predictor. In all the other steps, we include perceived nationality and speaker status as predictors. We would like to remind the reader that the common speaker was always Israeli. Therefore, the comparisons between uncommon and common speakers are between a common “Israeli” speaker and either an uncommon “Israeli” speaker, an uncommon “US‐originated” speaker, or an uncommon “FSU‐originated” speaker. We included this predictor because the effects we measure are relative, and are essentially compared between the speakers that each participant was exposed to. For example, in the alignment‐test step, the alignment effect is calculated as the difference between the probability of using the disfavored word with the “uncommon” speaker and the probability of using the disfavored word with the “common speaker.” Therefore, including this predictor enables us to evaluate the speaker‐specific effects, within the conditions of perceived nationality. All statistical analyses were done in R statistical software (R Core Team, 2023), using the RStudio environment (Posit team, 2023). The packages we used were the “tidyverse” (Wickham et al., 2019), “rlist” (Ren, 2021), “lookup” (Wright, 2021), “data.table” (Barrett et al., 2023), “interactions” (Long, 2021). “ggsignif” (Ahlmann‐Eltze & Patil, 2022), “car”(Fox & Weisberg, 2018), “emmeans” (Lenth, 2023), “ordbetareg” (Kubinec, 2022), “lme4” (Bates et al., 2015), and “lmerTest” (Kuznetsova, Brockhoff, Christensen, & Jensen, 2020). Where post‐hoc pairwise comparisons were made to interpret interactions, we used the “FDR” (Benjamini & Hochberg, 1995) method to correct for multiple comparisons. All mixed‐effects models included the maximal random structure supported by the data (Barr, Levy, Scheepers, & Tily, 2013; Matuschek, Kliegl, Vasishth, Baayen, & Bates, 2017) that did not introduce singularity (over‐fitting) issues. The random structure is detailed for each mixed‐effects model. All contrasts were sum‐coded. Data, materials, and analyses are available https://osf.io/zesj3/?view_only = 99505869c7b04fa9847fe9491de9a736.
Results
Detection
The ability to correctly map interspeaker variability was analyzed using the discriminability index (d’) of the Signal Detection Theory (Swets, W., & Birdsall, 1961). Using the “psycho” package (Makowski, 2018), we calculated d’ (Z(Hits) – Z(False Alarms)) for each participant. We tested each perceived nationality (IL, FSU, and US) separately to determine whether the group mean d’ was significantly different from 0 (one‐sample t‐test) to assess speaker‐specific mapping within each group. Additionally, we compared the mean d’ between the conditions of perceived nationality, using unpaired‐sample t‐tests, to assess mapping differences between the groups.
The d's in all perceived nationalities were significantly larger than 0 (IL: M = 2.93, SD = 0.77, t(31) = 21.58, p < .001; FSU: M = 3.25, SD = 0.52, t(31) = 35.15, p < .001; US: M = 2.97, SD = 1.00, t(31) = 16.86, p < .001; Fig. 2), indicating that in all perceived nationalities, participants correctly mapped each speaker to his/her word choices. Between‐group comparisons did not reveal any significant differences.
Alignment
We defined the alignment effect as the increased tendency to produce disfavored words to the speaker who used them in the first place than to the other speaker. Therefore, this analysis included only the disfavored words. A mixed‐effects logistic regression (using the binomial link function) model was fitted using the “lme4” package (Bates, Mächler, Bolker, & Walker, 2015), predicting the log‐odds of producing the disfavored alternative for each image by perceived nationality (IL/FSU/US) and by the language use status of the speaker (henceforth, speaker status; i.e., common/uncommon). For this sake, each instance of a disfavored word was coded as 1 and each instance of a favored one as 0. This model also included random intercepts and random slopes of block order by participants.
The mixed‐effects logistic regression model revealed a significant effect of speaker status (p < .001; see Table 1) and a significant perceived nationality by speaker status interaction (p = .01). The speaker status effect indicates that the probability of using a disfavored word is higher (Mean probability = 27.46%) when interacting with the uncommon speaker than when speaking to the common speaker (Mean probability = 13.99%; Odds Ratio = 3.52). This effect represents the alignment effect because it means that participants were more likely to use the disfavored words with the speaker who produced them before (the uncommon speaker), thereby aligning with them.
The significant interaction between perceived nationality and speaker status (p = .010; Table 1 and Fig. 3) indicated that the alignment effect differed between the perceived nationalities. Planned between‐group comparisons revealed that the alignment effect was significant under all perceived nationalities. However, the effect was significantly larger in the IL (Odds Ratio = 4.07, Z = 8.21; p < .001) and in the FSU (Odds Ratio = 4.47; Z = 8.62; p < .001) conditions than in the US condition (Odds Ratio = 2.41, Z = 6.07; p < .001). This difference in magnitude stemmed from the marginally significant increased tendency to use the disfavored words with the common speaker (i.e., the speaker who did not use them) in the US condition compared to the IL (Odds Ratio = 3.84; p = .054) and FSU (Odds Ratio = 5.08; p = .04) conditions. Additionally, following Branigan, Tosi, and Gillespie‐Smith (2016), we conducted an analysis evaluating the probability of using a disfavored word by perceived nationality and speaker status relative to the baseline probabilities obtained from the norming study in Trainin and Shetreet (2025). This analysis showed that only when the uncommon speaker was perceived as American, the probability of using a disfavored word was higher than under baseline conditions (OR = 1.595 for the American speaker and OR < 1 for the rest; for the full analysis, see Supplementary Materials).
Linguistic generalization
This analysis included two steps. The first step followed the analysis conducted in Trainin and Shetreet (2025), assessing the effects of the factors manipulated in this speaker status and speaker's perceived nationality. The second step aimed to address the possible effects of the a priori bias on linguistic evaluations of foreign speakers. Thus, the analysis included a separate norming study conducted to evaluate linguistic expectations, based on speakers’ names alone, without engaging in any kind of interaction with them, and, therefore, without being exposed to their language use.
Linguistic expectations after exposure to language use
We analyzed the responses for each linguistic phenomenon separately, by fitting three separate mixed‐effects logistic regression models with perceived nationality (IL/FSU/US) and speaker status (common/uncommon) as fixed effects. The models also included random intercepts by participants. These models predicted the log‐odds of responding “yes” to the association of the disfavored utterance of each phenomenon (disfavored constituent order, disfavored adjective order, and new disfavored words) with a given speaker. In the lexical items model, there was zero variance in the response (complete separation; (Albert & Anderson, 1984) when the speaker was the uncommon speaker in the IL condition, making the model uninterpretable. Following Clark, Blanchard, Hui, Tian, and Woods (2023), we fitted a Penalized Logistic Regression (Firth, 1993) model, using the “blme” package (Chung, Rabe‐Hesketh, Dorie, Gelman, & Liu, 2013). This method of model fitting shrinks the model coefficients in a way that makes them usable.
The models testing the phrase‐level items did not reveal any significant effects. The model testing sentence‐level items revealed a marginally significant speaker status effect (p = .0498; Table 2), such that the uncommon speaker was perceived—across perceived nationalities—as more likely to have produced a disfavored constituent order. However, planned pairwise comparisons ^2^ revealed that this effect was significant in the FSU condition (Odds Ratio = 5.00, Z = 2.78, p = .005), but not in the IL (Odds Ratio = 1.52, Z = 0.79, p = .43) and the US (Odds Ratio = 1.00, Z = 0.00, p = 1.00) conditions (Fig. 4). The lexical items model revealed a significant speaker status effect, such that the probability of a positive response was higher for the uncommon speaker than for the common speaker across all perceived nationalities (IL: 100%; US: 81.25%; FSU: 90.625%). Planned post‐hoc pairwise comparisons revealed that this effect was comparable across all perceived nationalities (Odds Ratios > 10^7^; ps < .001).
The results from the lexical items suggest that, regardless of perceived foreignness, listeners generalize the tendency to use disfavored words, at least in cases like the current study where ample evidence for this tendency was given (i.e., a set of highly disfavored words was used repeatedly and frequently). However, the results from the sentence‐level model suggest that linguistic generalization is conditional on the identity of the speaker. When the speaker had a Russian name, participants assumed that his/her use of disfavored words implied the use of disfavored constituent ordering. When the speaker had an Israeli or an American name, participants did not consider the use of disfavored words as sufficient evidence for overall disfavored language use.
Linguistic expectations without exposure to language use
To evaluate a priori linguistic expectations of speakers from the different perceived nationalities in our study, we fitted mixed‐effects logistic regressions for each linguistic phenomenon separately, predicting the log‐odds of a positive response by perceived nationality (IL/FSU/US). In cases where the model fitting resulted in singularity issues, we used penalized models with the “blme” package (see Section 3.4.1).
In the norming study, the disfavored adjective ordering items showed no effect of perceived nationality (ps > .38). In the constituent order items, no statistically significant effects were observed (ps = .128). However, interestingly, in the lexical items, there was a significant effect of perceived nationality, such that the FSU speakers were much less expected to use the disfavored words than the IL speakers (OR = 5.150, p = .039) and the US speakers (OR = 10.098, p = .016). This suggests that a priori, native Hebrew Israeli speakers are more likely to expect the use of disfavored words from an American (85%) or Israeli speaker (75%) than from an FSU immigrant (40%).
With versus without exposure to language use
The norming study also allowed us to compare the expectations about speakers based on their name alone with the updated expectations after exposure to their language use. For each linguistic phenomenon, we fitted a mixed‐effects logistic regression model predicting ratings by exposure to language use (with/without) and perceived nationality (IL/FSU/US). We then performed planned post‐hoc pairwise comparisons within each condition of perceived nationality between the experiments.
This analysis revealed that the expectations toward IL and US speakers, regarding all the linguistic phenomena, did not differ between participants who were exposed to their language use and those who were not (ps > .68), except a marginal trend of increased probability of a positive response in the IL condition for the lexical items (OR = 1.44, p = .063). In the FSU condition, the expectations toward the use of disfavored adjective or constituent orders were also unaffected by exposure to language use (ps > .24) (Figs. 5 and 6). However, for the lexical items, there was a significant increase in the probability of a positive response (OR = 22.6, p = .002). This suggests that the use of disfavored words by the FSU speakers in the experiment created a shift in expectations about them. Interestingly, while the expectations regarding the use of constituent orders differed for the FSU speakers compared to the IL and US speakers (when tested after exposure to language use), they were similar for FSU speakers with and without exposure to language use. This suggests that the use of this sentence form is a priori expected for FSU speakers, rather than formed by their word choices.
Social generalization
This analysis followed the steps of the linguistic generalizations, looking at the results of the main experiment, as well as those of a separate norming study (without linguistic interaction). The first step followed the analysis done in Trainin and Shetreet (2025), assessing the effects of the factors manipulated in this speaker status and speaker nationality. The second step addressed the possible effects of the a priori bias on social evaluations of foreign speakers.
Social perception after exposure to language use
We fitted four separate ordered beta regressions (Kubinec, 2022), one per question. Each model predicted the numeric rating by perceived nationality (IL/FSU/US) and speaker status (common/uncommon), as well as the interaction between them. All models included random intercepts by participants. Unlike the previous analyses we reported, this one employs a Bayesian framework. Thus, in this analysis, there is no clear significance threshold. Nevertheless, in this paper, we adopt a semi‐frequentist way of interpreting Bayesian models, which means that when the coefficients’ 95% credible intervals do not include 0, we are adequately confident that a significant difference exists. Thus, although the Bayesian framework does not offer “significance tests,” we use “significance tests” terminology in the section for consistency with the rest of this paper. Additionally, because we were interested in all the pairwise comparisons, we directly report the results of these planned pairwise contrasts (see full model summaries in https://osf.io/zesj3/?view_only=99505869c7b04fa9847fe9491de9a736.) ^3^
We first report the results of the “linguistic competence” questions (born abroad, reading books), where two patterns emerged. For the book reading question, pairwise contrasts revealed that across all perceived nationalities, the uncommon speaker was perceived as having read more books than the common speaker. This means that the speaker who used more disfavored words, whether foreign or not, was perceived as more literate (Table 3 and Fig. 7). This may be because most disfavored words are likely in a higher register. For the Foreignness questions, the pairwise contrasts revealed that in the IL condition, as expected, neither speaker (both having Israeli names) was estimated as likely to have been born abroad. In the FSU condition, where the uncommon speaker had a Russian name, participants estimated that the uncommon speaker was more likely to have been born abroad than the common speaker, as expected. A surprising result was demonstrated in the US condition, in which the uncommon speaker (with an American name) was not assessed as more likely to have been born abroad than the common speaker. This raises the possibility that the “American” names we included were not perceived as actually American by participants. We address this possibility in the following norming study, where we show that this is unlikely. Thus, the perceived foreignness of the American speakers shows a language use effect. We discuss this further in the General Discussion (Section 4.1).
Concerning the “sociability” questions, the pairwise contrasts of the number‐of‐friends question revealed that the uncommon speaker was perceived as having fewer friends, across all perceived nationalities. However, when contrasting the ratings of the uncommon speaker between perceived nationalities, the uncommon speaker in the US condition was perceived as having more friends than in both the IL condition (HPD Interval = 0.111–0.334) and the FSU condition (HPD interval = 0.021–0.250). This indicates that although the “American” speakers were socially “punished” for using disfavored words, they paid a smaller price than the uncommon “Israeli” or “Russian” speakers. For the cooperativeness question, pairwise contrasts revealed that the uncommon speaker was perceived as less cooperative than the common speaker in the IL and the FSU conditions, but not in the US condition. This again indicates that participants were more lenient toward the “American” speakers and were less likely to judge their use of disfavored words as a breach of cooperativeness.
Social perception without exposure to language use
As in the analysis of the data from Step 5 in the main experiment, we fitted mixed‐effects ordered beta regression models (Kubinec, 2022) for each of the questions on the norming study. The questions we included in the norming study were the ones we included in the main experiment, except for the cooperativeness question, which would be considered infelicitous. The results of the norming study are summarized in Table 4 and Fig. 8.
As expected, the Israeli characters were perceived as less likely than the FSU‐originated and US‐originated characters to have been born abroad. There was no significant difference between the FSU and the US conditions in this question. The number‐of‐friends and number‐of‐books questions did not reveal any significant difference. Importantly, the foreignness question allowed us to conclude that our experimental manipulation effectively highlighted the foreignness of the uncommon speakers in the FSU and the US conditions, despite using a weak cue (names only). That is, when introduced to the speakers with those perceived nationalities, it is most likely that participants did in fact view both of them as foreign. Somewhat surprisingly, despite the differences in stereotypes between FSU and US immigrants, no difference was observed between the speakers in perceived friendliness or literacy. This could be due to the relatively weak foreignness signals, which consisted of different names only (without differences in appearance or accent). We elaborate on that in the General Discussion.
With versus without exposure to language use
The norming study also allowed us to provide interesting insights into the perception of foreign speakers with and without exposure to language use, by comparing the ratings from the norming study and the main experiment. For each question, we fitted an ordered beta regression model predicting ratings by exposure to language use (with/without) and perceived nationality (IL/FSU/US). We then performed planned post‐hoc pairwise comparisons within each condition of perceived nationality between experiments.
For the number‐of‐books question, ratings were higher in the main experiment, under all perceived nationalities (Table 5 and Fig. 9). This is expected, given that participants in the main experiment were exposed to language use that probably included words in a higher register. For the foreignness question, we observed an intriguing pattern, such that exposure to language use reduced the likelihood that the American speakers were perceived as having been born abroad (HPD Interval = 0.117–0.610), but it did not have a reliable effect on this likelihood for the FSU (HPD Interval = −0.116 to 0.343) or Israeli speakers (HPD Interval = −0.197 to 0.312). Similarly, both the FSU (HPD Interval = −0.092 to 0.328) and the Israeli speakers (HPD Interval = −0.092 to 0.328) were perceived as having fewer friends after being exposed to their language use compared to without exposure. This effect was virtually absent in the US condition. Thus, overall, we see a more positive evaluation of the American speakers. We discuss this in more detail in the General Discussion.
General discussion
In this study, we examined the effect of perceived foreignness on the ability to map interspeaker variability, on the extent of lexical alignment, and the nature and extent of generalizations made based on speakers’ language use, both within the linguistic domain and the social domain. Perceived foreignness, as determined by the name of the speaker, did not interfere with the mapping process, as reflected in the detection‐test step. However, perceived foreignness did have significant effects on the consequences of the mapping between speakers and their language use, including lexical alignment, linguistic generalization, and social generalization. First, we observed a diminished alignment effect in the US condition compared to the IL and FSU conditions (stemming from the overall increased tendency to use the disfavored words with both speakers in the US condition). Second, the use of disfavored words shifted expectations toward the use of disfavored words, compared to the a priori expectations toward the FSU speakers, but not toward the speakers from the other nationalities. Finally, group‐associated stereotypes affected these generalizations differently. While “Israelis” and “Russians” using disfavored words were perceived as less sociable than their common counterparts, “Americans” were not. Further, after being exposed to their language use, the uncommon “American” speakers gained more linguistic competence points than the “Russian” ones (as reflected in the perceived foreignness question).
Social meaning
Recent proposals (e.g., Beltrama, 2020; Beltrama, Solt, & Burnett, 2022; Kuperwasser & Shetreet,submitted) suggest that social meaning operates in two constructing social information through linguistic features (Eckert, 2012) and processing language based on social features. Our findings highlight not only that these two directions occur in the context of word selection, but also their reciprocal bidirectional nature, as both directions interact and influence one another in shaping social meaning. Prima facie, it seems that our findings can be categorized in one direction. For example, the social generalizations tested in the last phase of the experiment reflect the traditional direction, from language to social information, whereas the alignment effect or the linguistic generalizations reflect the complementing direction, from social information to language processing. However, we see differential effects based on the nationality associated with the speaker's name. This, therefore, suggests that the relationship between learning about a person based on their language use and the linguistic expectations that are formed toward that person based on their identity is more dynamic than this dichotomic categorization.
As mentioned above, social generalizations can be linked to the construction of social information from language use, the traditional direction of social meaning. Within this frame, language use (in our case, naming preferences) serves as a marker for social traits. Traditionally, these markers were examined with respect to social group membership (e.g., Tobar‐Henríquez et al., 2021), such that some forms of language use are indicative of belonging to certain social groups. Our results provide evidence that interlocutors gather social information concerning individual speakers from the way they use language (see also Beltrama & Papafragou, 2023). That is, the use of disfavored names for objects acts as a social marker at the individual level, and not merely as a proxy for traits associated with group membership. Replicating results from Trainin and Shetreet (2025), we showed that the uncommon speaker—the speaker who uses disfavored words—was seen as more literate. However, other dimensions of the social generalizations were modulated by social information about the in the IL and the FSU conditions, the uncommon speaker was perceived as having fewer friends and as less cooperative, but this was not the case in the US condition. This differential pattern suggests that the effect of language use on an individual's social perception depends on the speaker's identity.
The direct effect of language use should be seen in the comparison between the norming study, which introduced the speakers with no language exposure, and the main experiment, where participants were exposed to the language use of the speakers. For example, “American” speakers were perceived as far less likely to be foreign in the main experiment, but not in the norming study. That is, using disfavored words made the “American” speakers sound native. Yet, the norming study provides further evidence for this bidirectional relationship, as this effect was not found for the “Russian” speakers, although their use of disfavored words far exceeded a priori expectations. Furthermore, differential effects were seen in the number‐of‐friends question, where lower ratings in the main experiment compared to the norming study were seen in the IL and FSU conditions but not in the US condition. These findings suggest that social generalizations based on language use are dependent on a priori expectations regarding speakers.
Our study also provides evidence for the other direction, from social information to language processing, recently suggested as a component of social meaning. As we stated above, the social identity of the speaker moderated the alignment effect and linguistic generalizations. However, both cases demonstrate a link to specific social characteristics rather than to foreignness as a general social construct. Specifically, although all conditions of perceived nationality showed an alignment effect, its strength was modulated in the US condition, with a smaller alignment effect in the US condition than in the FSU and IL conditions, and a certain linguistic generalization was seen only in the FSU condition (VSO sentences), regardless of exposure to language use.
Let us first consider the long‐standing finding of speaker‐specific lexical alignment (Brennan & Clark, 1996; Garrod & Anderson, 1987), replicated here. While the alignment effect was evident under all experimental conditions, it was smaller in the US condition. This difference in the effect's magnitude mostly stemmed from the greater likelihood of participants to use the disfavored words with the common speaker in the US condition than in the IL and the FSU conditions. Note that across all perceived nationalities, the common speaker had an Israeli name, representing a native speaker. If so, the difference between the conditions does not come from the common speaker, but rather from the status of the uncommon speaker and its effect on the status of the words they used. In other words, participants more readily used disfavored words with a speaker who did not originally use these words, when their other partner was perceived as American. This suggests that the disfavored words became more acceptable and accessible due to their use by an American compared with the baseline conditions. A possible reason for this is the positive social status of Americans in the Israeli culture (see our discussion of stereotypes in interaction below).
We remind the reader that social generalizations were modulated by the speaker's perceived nationality (in the main experiment, but not in the norming study). That is, it is not only that people from different nationalities were socially perceived differently, but their language use was taken as differential evidence for their social profile. With that in mind, it is possible that the different alignment effect in the US condition has caused disfavored language use to be less salient (or even less disfavored). This, in turn, might have served as a weaker cue for social generalizations. In other words, because the social generalizations are based on the use of disfavored words, they might be less dramatic when the words’ preference status is upgraded. Interestingly, in Trainin and Shetreet (2025), the degree of words’ preference moderated the social generalizations, such that an uncommon speaker who used slightly disfavored words (in a ratio of approximately 40) was not considered socially incompetent to the same extent as a speaker who used highly disfavored words (in a ratio of approximately 20, as in the current study). Furthermore, the probability of aligning with the former was higher. Therefore, the varying effects in the social generalizations could stem from differing perceptions of words’ preference in the different perceived nationalities.
The second case of social features affecting language processing is seen in the linguistic generalizations, where we found that the uncommon speaker perceived as Russian, and not as Israeli or American, was rated as more likely than the common speaker to use disfavored sentence orders. Given that this was also true in the norming study, our results suggest that Russian immigrants are generally expected to use disfavored sentence structures, regardless of their word choices.
Interestingly, something else occurs at the word level. In the norming study, FSU speakers were less expected than the IL and the US speakers to produce disfavored words. This suggests that a default stereotype is that FSU speakers adhere to more normative or “proper” language use. However, after exposure to the speakers’ language use, FSU speakers were substantially more expected to use the disfavored words than under baseline terms, whereas in the IL condition, this trend was smaller, and in the US condition, no such effect was observed. This demonstrates the dynamic interplay between social stereotypes and linguistic input, as baseline expectations shifted according to language use in FSU speakers. This indicates that stereotypes, while present, are not listeners are willing to revise their beliefs concerning the speaker, based on direct linguistic evidence. In contrast, the absence of a similar shift for Israeli‐ and American‐named speakers suggests that participants’ baseline expectations for these groups already aligned with their linguistic behavior. Together, these results highlight the asymmetry in how social meaning and language use inform one while social categories shape initial assumptions about language, actual language use can recalibrate those assumptions. Critically, this recalibration depends on the speaker's social identity and the strength of the associated stereotype. In other (Bayesian) words, participants seem to have different prior beliefs regarding language use based on social characteristics of the speaker, which affect the generalizability of their linguistic output.
All in all, the bidirectionality of social meaning can be explained using Bayesian terminology. In this terminology, there are some priors that reflect initial beliefs or assumptions about a certain parameter, which are updated based on experiences or new systematic information to create posterior beliefs. In our study, the names of the speakers created certain “social priors” associated with the speaker's perceived nationality. Then, with evidence from the language use, these social impressions shifted and created new posterior distributions. Additionally, previous research (Trainin & Shetreet, 2025) further assumes some “linguistic priors,” such as the conditional probability of a word being used to describe a given object. The linguistic priors could be updated based on the speaker's language use, based on the speaker's identity, or based on the combination of these two factors. In the US condition, the “linguistic” priors were quite dramatically updated, as seen in the alignment effect (driven by the production of disfavored words to the common speaker). As discussed above, it is possible that the update was driven by the association between the positively perceived American identity and the disfavored words. This positive link can be seen as driving a dramatic update to the beliefs regarding these words, increasing the likelihood of them being used, and reducing their salience as a social marker, thereby making them a less dramatic ground for updating social features.
At this point, it is worth noting that the social priors we discuss may be weak (or less informative) in the context of our study, because they were based on the characters’ names only, and did not include other informative features such as accent or appearance. While this approach allowed for controlled manipulation of perceived nationality, which was found effective (as seen in the social norming study), it might not fully capture the complexities of real‐world social interactions. For example, it has been shown that accent has a substantial effect on speakers’ perception (Lev‐Ari & Keysar, 2010). Indeed, in our study, the social norming study did not reveal substantial differences between speakers based on their names alone (other than their foreignness status). Only when additional information (in the form of language use) was made available, some differences were observed, suggesting that social generalizations are based on accumulating evidence. The stronger the evidence, the stronger the generalizations. Therefore, the effects we observed could have been much stronger (or more complex) with auditory stimuli. Future research could use more naturalistic methods to investigate these phenomena.
Perceived foreignness and stereotypes in interaction
The above results demonstrate differential effects for different foreign groups, those associated with negative stereotypes (FSU) and those associated with positive stereotypes (US). Interestingly, the overall pattern of results was similar in the FSU and IL groups. This similarity may come from the fact that despite having suffered from social prejudice throughout the years, FSU immigrants are generally expected to be integrated into Israeli society (Cohen & Prashizky, 2024). This expectation may arise because the vast majority of FSU immigrants arrived in Israel in the early 1990s, approximately 30 years prior to this study. Yet, “Russian” speakers were a priori perceived as more likely to use a disfavored sentence order. This effect may arise from the negative stereotype against FSU immigrants, which sees them as separatist or condescending. Because the generalization to disfavored language in the sentence level also probably involved a higher register, it may convey distance from the common speech of the community, aligning with the perception of separatism. The finding that “Russian” speakers were less likely, a priori, to use disfavored words, may stem from the expectation that their Hebrew lexicon would be smaller, perhaps due to the assumption that they do not make an effort to learn a broader lexicon in Hebrew. Concerning the social generalizations, we would like to note that, compared to the “American” speakers, they received harsher judgments (in terms of cooperation and the number of friends). This could simply be related to the negative stereotypes associated with FSU immigrants. However, it is possible that their use of disfavored words was seen as a greater violation of social expectations (as is also evident from the linguistic expectations norming study), leading to stronger negative reactions than those aimed at Americans, who may not face the same integration expectations. This latter explanation seems even more plausible when the results of the norming study are taken into account. Without exposure to (disfavored) language use, the “Russian” and the “American” speakers were not perceived differently in social terms, as one might expect due to the different a priori stereotypes. However, the speakers’ language use led to differential shifts in how they were socially perceived, suggesting that the updating process of the social perception, instead of the initial perception, depends on these stereotypes. Considering the linguistic generalizations versus linguistic expectations together with the social generalizations, the FSU speakers were expected to have smaller lexicons in Hebrew and were not initially perceived as less sociable. However, once it turned out they were familiar with uncommon Hebrew words, it is possible that they were expected by (Israeli) participants to have integrated more fully into Israeli society, and were, therefore, seen as less sociable. For the US speakers, however, the linguistic expectations were somewhat looser, and the social price they paid was minimal.
Indeed, favorable responses were seen for the “American” speakers. This more positive treatment of “American” speakers may be due to the generally favorable view of Americans in Israeli society, where being American is often considered socially desirable or “cool” (Avraham & First, 2003; Rebhun & Waxman, 2000). As a result, deviations from linguistic norms by Americans might be viewed more leniently or even positively. In fact, this may also have led participants to regard the “American” speaker's naming preferences as less disfavored, and use them themselves even with a speaker who did not produce these words.
That is, although all speakers used disfavored words, listeners tended to treat the “American” speakers as more socially competent than their “Israeli” and “Russian” counterparts. This suggests that social generalizations of speakers based on their language use are conditional on the a priori perception of their social group. More specifically, this suggests that the a priori perception of Americans as more socially competent ameliorated the social “punishment” for using language in a disfavored fashion.
These findings highlight the importance of a priori knowledge regarding a person (as part of a social group or not) in determining how their language use will be perceived and generalized. Moreover, the findings suggest that the social penalties for using disfavored language depend on whether a group is expected to integrate more fully into local mainstream norms. When such integration is expected, as with FSU immigrants, failing to conform to preferred forms of language use may be “punished” more severely. On the other hand, at least in Americanized societies such as the Israeli one, American immigrants are not met with the same integration expectations, nor are they facing the consequences of being outsiders to the same extent that immigrants from less “desirable” countries are facing. This could serve as a reminder that integration into a community depends on both the native and the non‐native groups.
Finally, these results point to the potential benefit of teaching immigrants not only the “correct” forms of language but also the socially favored ones. Knowing all the grammatically acceptable forms is not enough—learning the socially normative language patterns can be crucial for smoother social integration. Emphasizing these subtle but important distinctions could help immigrants navigate the social landscape more effectively and avoid unnecessary social penalties for language use that, while correct, might not be socially acceptable in certain contexts. With this in mind, it could also be valuable to evaluate how teaching the socially appropriate ways to communicate in a second language affects social perception.
Funding
This study was supported by the Israel Science Foundation (ISF) grant number 811/23 to ES.
Conflict of interest
The authors have no conflict of interest to disclose.
Ethical statement
This study was approved by Tel Aviv University's ethics committee (approval number 0005280‐1).