Native English speakers’ structural alignment mediated by foreign-accented speech

1 Introduction

The majority of the world’s population is bilingual or multilingual, so it is very common for interlocutors to have different first languages. In addition, the rise of technology, globalization, and interconnectedness between populations have produced variations in languages, mostly English, as it is the most widely spoken language. This leads even monolingual English speakers who have no experience with another language exposed to language variation, specifically in the form of foreign accent that they must accommodate to. How speakers deal with language variation is a topic of debate among researchers in many fields but very little is known about how monolingual English speakers accommodate variation from speakers with foreign-accents, or a type of English that is deviant from their own accent. This provided motivation for the current study. Present theories of processing foreign-accented speech lack unanimity and disregard the effect of social factors. Understanding how monolingual English speakers respond or react to non-native speech is important for models of language acquisition, comprehension, and production.

One well-tested measure of response to speaker variability is adaptation or alignment. This can be manifested as linguistic priming, where the exposure to one stimulus at a certain linguistic level influences the response of another stimulus at the same linguistic level. The current study employed a syntactic priming paradigm, when a speaker is more likely to produce an utterance using the same syntactic structure they have just recently heard or used (Bock 1986). On prime trials in a typical syntactic priming study, participants hear and repeat a prime sentence in a particular syntactic form. This can be a manipulation of the dative structure (prepositional object (PO) construction; e. g., “The girl gives the apple to the boy”, or double object (DO) construction; e. g., “The girl gives the boy the apple”), or the transitive structure (active; e. g., “Lightning is striking the church” or passive; e. g., “The church is being struck by lightning”). Then, participants are shown a picture that depicts an event and are asked to describe the picture. It has been repeatedly found that participants are much more likely to repeat a structure they just heard when producing their own original description. This kind of alignment has been shown at many different linguistic levels such as vowel production (Abrego-Collier et al. 2011), concepts and mental models (Brennan and Clarke 1996), and referring expressions (Garrod and Doherty 1994).

Theories in psycholinguistics have attempted to conceptualize alignment through processes such as Implicit Learning (Bock and Griffin 2000; Chang et al. 2006) which is a cognitive mechanism that undergoes long term adaptation through comprehension of stimuli. On the other hand, Residual Activation theory claims alignment is achieved through temporary activation of mental representations in memory that correspond to information recently used. According to Pickering and Branigan (1998), processing a sentence activates lexical and combinatorial nodes in memory. The residual activation of these nodes makes it more likely that the same structure is produced or expected when processing the following (target) sentence. These theories operate under the assumption that alignment is an automatic cognitive process to aid communication. However, they do not address social factors that could influence alignment even though previous research have given evidence for the effects of various social elements on language processing. The Interactive Alignment account (Pickering and Garrod 2004) touches on the role of the interlocutors and social factors by assuming that alignment of various representations between speakers is important for successful dialogue. This model assumes a link between perception and behavior, and addresses the priming of various levels of representations between the listener and the speaker to achieve common ground. The interactions and shared mental models between two speakers is the driving force behind linguistic alignment. However, this model does not go into detail of the specific social factors that can influence alignment.

Recently, various social factors have been found to affect alignment at different linguistic levels: language distance (Kim et al. 2011), attitude (Abrego-Collier et al. 2011), perceived smartness of the speaker (Weatherholtz et al. 2014), prestige of the speaker (Lev-Ari 2015) and disposition as liking or disliking the interlocutor (Balcetis and Dale 2005). However, most of the previous work investigating these social roles have focused on phonetic alignment, such as vowel convergence between college roommates (Pardo et al. 2012), extended VOT mimicry as a function of attitude (Abrego-Collier et al. 2011), phonetic convergence between native and nonnative speakers (Kim et al. 2011), and accent alignment in dialectical interactions (Babel 2010).

In contrast to phonetic alignment, syntactic alignment in relation to social factors has been rarely studied. Balcetis and Dale (2005) investigated syntactic alignment by manipulating the interpersonal relationships between interlocutors. Participants conversed with a confederate who was either nice or mean, and found greater syntactic alignment for active, passive, and PO structures with the nice interlocutor compared to the mean interlocutor. Results from this study show that interpersonal disposition between speakers can influence linguistic alignment at the syntactic level as well. Lev-Ari (2015) examined the role of prestige, similarity, and liking of the interlocutor in grammatical convergence of Dutch speakers. Participants were asked to listen to a passage in which either a subject-verb or verb-subject order was used. Participants were told that the recorded message was from a student in the top 10 % or bottom 10 % (manipulating the prestige), or from a student that performed better than them, similar to them, or worse than them (manipulating similarity). Syntactic alignment was tested by having participants unscramble sentences. At the end of the task, participants rated how much they liked the speaker. Results show that grammatical alignment depends on social status and how much the individual is liked. Alignment is contextual and can change from one situation or individual to another. In addition, Weatherholtz et al. (2014) explored the effect of the perceived standardness of speakers’ accent on syntactic alignment. They used various social dimensions as predictors for the degree of dative syntactic alignment. Participants took part in a picture description task after being primed by a passage containing dative sentences with heavy ideological opinions of government spending practices. The passage was recorded with a Standard English accent, an African-American English accent, and a Chinese-English accent. Participants also completed a questionnaire assessing multiple social dimensions. According to their results, general priming was automatic, occurring across all speaker conditions but the degree of syntactic alignment was mediated by a variety of factors including a perceived standardness of the speaker’s accent, perceived smartness of the speaker, perceived similarity to the speaker, and preference for compromise as a conflict management style.

These studies began uncovering the importance of social aspects in language contexts. In line with these studies, the current study aims to provide more evidence for the effect of social factors on syntactic alignment. Specifically, we investigated whether native American English speakers show different syntactic alignment to native American English versus foreign-accented English (here defined as a non-American English accent). Our study differs from Weatherholtz et al.’s in various ways. First, Weatherholtz et al. did not investigate participants’ familiarity with different accents of English which could in turn affect how standard they perceive the accents. In the current study, we further explore if familiarity with each specific accent is the driving force behind different patterns of alignment since familiarity could play an important role in the processing of foreign accent. It has been found that a listener’s awareness and interactions with different varieties can change the perception of foreign accented speech. Kraut and Wulff (2013) found that participants who self-reported low degrees of familiarity with foreign accented speech rated foreign speech samples to be less comprehensible and having lower communicative ability than the raters who were more familiar with foreign accents. These findings suggest that familiarity with a foreign accent can alter the social perception of that foreign accent.

Another difference between the present study and Weatherholtz et al.’s is that the latter used a between – subject design where participants only heard one type of structure prime by one accented speaker. Furthermore, Weatherholtz et al. did not investigate trial-to-trial priming but presented the prime in an exposure block. Weatherholtz et al. therefore did not investigate how alignment can change within an individual in response to different accents or structures.

Therefore, the present study employed a within-subject, immediate priming design to investigate how native English speakers align with various foreign-accented speech at the syntactic level and to what extent familiarity with an accent can play a role in the processing of alignment responding to foreign accented speech. We predicted that native speakers of English would show greater syntactic alignment with the native American speaker accent condition than the foreign accent conditions, and that priming is stronger for the accents rated as more familiar.

2 Method

2.3 Procedure

Stimulus presentation and data collection in the priming study was controlled by E-prime 2.0 professional. Figure 1 depicts the structure of a trial. Each trial started with a fixation presented for 100 ms. Then, a prime picture (e. g., “a helmet”) appeared, and the prime sentence was played. The participants were instructed to listen to a prime sentence (e. g., “The policeman gives a helmet to the driver”) and repeat it. Next, they were asked to indicate whether the prime picture was related to the prime sentence by pressing 1 for “YES” and 2 for “NO”. Subsequently, the participants saw a target picture (e. g., A man giving a present to the girl) and described the picture using the sentence fragment (e. g., “The man gives…”) given below the picture. The participants controlled their own pace by pressing the spacebar as soon as they finished describing the target picture. After precise instructions about the experiment, the participants practiced the task with transitive practice trials consisting of 3 Prime-Target pairs until they were fully aware of how to perform the task. None of the practice trials included a dative construction.

Figure 1:

An example of a Prime-Target pair.

The three speaker conditions (American speaker, Korean speaker, Indian speaker) were blocked and the order of the speaker was randomized across the participants. After each speaker condition, the participants were asked 1) what kind of accent they thought the speaker had, 2) how familiar they were with that accent they heard. They rated the degree of their familiarity with the accent on a scale from 1 to 7, with ‘1′ being not familiar at all and ‘7′ being very familiar.

3 Results

First, we checked whether the speaker conditions were well manipulated enough for participants to perceive the three speakers’ accents differently. All participants correctly identified the American speaker and the Indian speaker as coming from the US and India, respectively. However, they could not always correctly identify the Korean speaker, and often responded that the speaker came from an Asian country. The participants’ rating on familiarity showed that the familiarity with each accent was successfully manipulated; the American speaker was rated with a mean of 6.96 (SD=0.27), the Indian speaker with a mean of 3.73 (SD=1.58), and the Korean speaker with a mean of 3.15 (SD=1.51). The participants rated the American speaker as more familiar than each of the non-American speakers (American speaker vs. Korean speaker: t(42)=16.02, p < 0.0001, American speaker vs. Indian speaker: t(42)=12.94, p < 0.0001); and were less familiar with the accent of the Korean speaker than of the Indian speaker (t(42)=–2.33, p=0.02).

Only trials in which the prime was correctly repeated were included in the analysis. This ensured that participants had correctly understood the prime sentence, regardless of the speaker’s accent. Omitting the incorrect trials resulted in 2,230 responses (about 4 % data loss). Table 1 shows an overview of response types for each speaker condition. Native English speakers had a slight preference for the PO construction (See responses in the Baseline Prime condition). Numerically, the participants produced more DOs after a DO prime than after a PO prime, and more POs after a PO prime than after a DO prime.

We excluded OTHER responses from the analysis, as they were not related to the priming effect. Our final data set included 2,070 data points. Data were analyzed using a logistic mixed-effects model (R version 3.2.4: R Core Team 2016). The dependent variable was the number of PO responses (PO coded as 1 and DO coded as 0). Since we had three speaker conditions, we used a Helmert contrast coding for the factor Speaker Type, thus creating two contrasts: (1) American versus Non-American (the latter coded negative); (2) Korean vs. Indian (the latter coded negative). Prime Type was treatment coded (DO vs. Baseline, PO vs. Baseline, Baseline coded as ‘0’). We used treatment rather than sum coding for this factor, since we were particularly interested in the difference between the Baseline and DO prime (DO priming effect) and between the Baseline and PO prime (PO priming effect). Since the participants were exposed to each speaker condition in a blocked design, the order of exposure to speaker type can influence the results. To control for speaker order effect, we included the factor Speaker Order (sum-coded). The full model with the three fixed effect factors, all possible interactions of the fixed effect factors and maximum random effect structure did not converge. The model reported below included all three fixed effects factors (i. e., Prime Type, Speaker Type and Speaker Order), the interaction between Prime Type and Speaker Type, and the interaction between Prime Type and Speaker Order as fixed factors. As for random effects, we included intercepts for Participants and Items and a slope for Prime Type for both Participants and Items. Models with a more complex random effects structure or with interactions between Speaker Order and Speaker Type as fixed effects factors did not converge (Barr et al. 2013).

Results from the statistical analysis are given in Table 2. The participants revealed a robust priming effect for both DO prime (vs. Baseline) (β=–1.27, SE=0.30, z=–4.26, p < 0.0001) and PO prime (vs. Baseline) (β=1.88, SE=0.29, z=6.56, p < 0.0001). In addition, an interaction between PO prime (vs. Baseline) and Non-American speakers vs. American speaker was significant (β=–0.99, SE=0.45, z=–2.18, p=0.03), indicating that the PO priming effect (vs. Baseline) was smaller in the American speaker condition than in the non-American speaker conditions.

To further explore the effect of familiarity with each speaker’s accent on priming, we constructed a separate logistic mixed effects model including Familiarity as a fixed effect factor instead of Speaker Type. The other fixed effect factors and the random effect factors mentioned above remained the same. We centered Familiarity on the mean to reduce collinearity. As shown in Table 3, this model again yielded strong priming effects for both DO and PO prime conditions. There was also a main effect of Familiarity on PO production in the Baseline condition: the greater the familiarity with the speakers’ accents, the more production of the PO construction (See Figure 2). Additionally, the interaction between PO prime condition and Familiarity approached towards significance, suggesting that the PO priming effect is reduced as the familiarity with the accent increases. This patterns with the results in our first analysis, where we found smaller PO priming in the American vs. Non-American speaker conditions.

Figure 2:

Mean PO production as a function of Familiarity and Prime type. DbO stands for DO.

4 Discussion

In contrast with our predictions, the results of the current study revealed smaller PO priming in the American speaker condition than in the non-American (i. e., foreign-accented) speaker conditions. A further analysis including the effect of familiarity on syntactic priming suggests that familiarity may be a driving factor for this pattern: participants showed less priming for the PO construction, the more familiar they were with the accent. Clearly, native American English speakers in our study were the most familiar with the native American speaker’s accent and were significantly less primed for the PO construction in the American speaker condition.

Our findings are not compatible with the previous findings from Balcetis and Dale’s (2005) study, in which native speakers of English showed more priming for PO constructions when they performed a picture description task with a nice confederate than with a mean confederate. Given the well-known relationship between familiarity and liking (Zajonc 2001), it is expected that participants would demonstrate more alignment in the native American speaker condition. However, familiarity with the native American speaker’s accent reduced PO priming in our study. The difference in findings may be due to the use of foreign accents in our study, which may have led to differences in attention. Participants may have allocated more attention to the utterances with foreign accents to correctly understand them. Meanwhile, they needed not pay as much attention to the native American speaker’s utterances since they would instantly recognize it as a very familiar accent, leading to a reduced priming effect. This warrants future studies since less is known about the effect of attention on syntactic alignment. It is also possible that the difference in results between the two studies is attributable to the nature of task types. In Balcetis and Dale (2005)’s study, the participants performed an interactive task with the confederate and thus their result could elucidate the aspect of social interaction related to interpersonal relationship. On the other hand, the current study employed a non-interactive or socially-impoverished task which could show more automatic socio-cognitive effect (e. g., speaker type or familiarity with a specific accent) on syntactic alignment, not mediated by social interaction with an interlocutor.

The current study is more easily compared to Weatherholtz et al. (2014)’s since both studies employed a non-interactive picture description task and included foreign-accented speech. In Weatherholtz et al.’s study, the participants listened to one of three speakers’ diatribe containing political ideology with a different accent (i. e., accentedness manipulated in terms of standardness). Among various social factors, perceived standardness of the speaker’s accent, perceived similarity to the speaker and conflict management style influenced syntactic alignment. Greater syntactic alignment was observed when the participants listened to a speaker who they perceived to have a standard accent. This seems to be at odds with our findings that priming was smaller in the native American-English speaker condition. While accent standardness and accent familiarity can be assumed to be related, one is not necessarily a straightforward function of the other. Therefore, the differences between these variables could produce differences in alignment. However, we cannot exclude the possibility that the different patterns of alignment could be attributed to the differences in methodologies. Our study looked at trial-to-trial priming using a within-subject design while Weatherholtz et al. used a between-subject design with multiple social factors included in each block.

More interestingly, Weatherholtz et al. found a significant negative interaction between prime construction type and interpersonal similarity. In particular, as the participants perceived greater interpersonal similarity with a speaker, they were less likely to align with the PO construction. This result can be linked to our finding that native American English participants were less primed for the PO construction in the American speaker condition. Weatherholtz et al. manipulated interpersonal similarity using political ideology, whereas the participants in our study could feel similar to the American speaker as they have similar accents. The question still remains why the PO construction is more responsive to the influence of social cues or factors than the DO construction.

The findings from the current study are consistent with Weatherholtz et al.’s in that syntactic alignment was somewhat automatic (i. e., overall syntactic alignment across the speaker type) and the magnitude thereof was mediated by social factors (speaker type or familiarity with a speaker’s accent). Unlike sociolinguistic studies that regarded alignment as a strategic process for social goals, both studies could find automatic syntactic priming influenced by social factors even in the context that does not require any social interaction or strategy. This finding is in line with the work in social psychology that uncovered subconscious aspects of social perception (Monin 2003; Zajonc 2001) and the work in phonetic alignment that revealed social influence in non-interactive tasks (Abrego-Collier et al. 2011; Babel 2010).

Alternatively, in our study, the smaller PO priming effect in the American speaker condition could be due in part to a ceiling effect. Notably, more PO constructions were produced in the Baseline condition as the participants’ familiarity with an accent increased. This may not have allowed more room for a priming effect in the PO prime condition. However, the fact that participants were more likely to produce the PO structure with a more familiarly-accented speaker provides support for the view that PO alignment is sensitive to social factors.

Albeit not clear yet why syntactic alignment of PO and DO constructions showed different sensitivity to social factors, our findings contribute to the current literature on syntactic alignment as we discovered that 1) the speaker who possesses linguistic similarity can influence syntactic alignment (e. g., similar accent) and 2) it is highly likely that perceived familiarity with a speaker’s accent could play a role in syntactic alignment not just in phonetic alignment.

5 Conclusion

In summary, only a few studies to date have explored the effect of social factors on syntactic alignment and found that linguistic alignment does not purely reflect cognitive aspects of linguistic processing. Rather, alignment involves much more complicated aspects of linguistic processing. For instance, the magnitude of automatic syntactic alignment has been shown to be affected by different social cues (Balcetis and Dale 2005; Lev-Ari 2015; Weatherholtz et al. 2014). Although they reveal complex patterns, those studies on syntactic alignment provided evidence that subconscious linguistic processing can be influenced by various social factors at the syntactic level as well as the phonetic level (Kraljic et al. 2008). The current study provided more evidence for this claim since we found native American English speakers’ automatic syntactic alignment can be mediated by their perception of different accents (i. e., foreign accents vs. non-foreign accents) and their familiarity with the accents.

Taken together, all of these findings imply the necessity of considering linguistic variation from speakers for language processing. Understanding the effect of social or pragmatic factors (e. g., speaker-specific cues on syntactic alignment) and building up theoretical accounts with integration of social influences are greatly important given that language is always used in a social context embracing various factors. Finally, it is also worth noting that this study showed the possibility of syntactic priming paradigm as a method to elucidate a socio-cognitive aspect of language use.