Language anxiety in the immigrant context: Sweaty palms?

Abstract

Aims and objectives:

This is a proof-of-concept study designed to evaluate the level of language anxiety among immigrants by assessing autonomic arousal associated with heritage language anxiety and majority language anxiety among three generations of Turkish immigrants in the Netherlands. It examines the possible relationship between physiology, bilingual speech, language background variables and language anxiety.

Design:

Two measures of electrodermal activity – skin conductance level (SCL) and skin conductance response (SCR) – were recorded during a video-retelling task in six experimental phases: baseline (2×), free (bilingual) mode (2×), monolingual heritage-language (Turkish) mode, and monolingual majority-language (Dutch) mode. Participants (n=30) ranked their level of language anxiety after the Turkish monolingual mode and Dutch monolingual mode. A Likert scale-based questionnaire was used to gather information on language background variables (i.e. age of acquisition, oral language proficiency and frequency of language use).

Findings:

Third-generation bilinguals, to a greater extent than first- and second-generation bilinguals, demonstrated greater autonomic arousal during the Turkish monolingual mode than during the Dutch one. Participants’ SCLs/SCRs in monolingual modes were strongly correlated with their self-reports on heritage and majority language anxiety. Higher levels of SCLs/SCRs in the Turkish monolingual mode were negatively correlated with participants’ Turkish oral proficiency levels and frequency of daily use of Turkish. The correlations between SCLs/SCRs in the Dutch monolingual mode and participants’ Dutch oral proficiency levels and frequency of daily use of Dutch, on the other hand, were low to non-significant. These findings suggest that language anxiety is also related to social and psychological factors, rather than only self-perceived low language proficiency.

Implications:

The outcome confirms the presence of language anxiety in immigrant contexts. An interdisciplinary approach that applies physiological measures together with social factors and self-reports can shed further light on language anxiety.

Originality:

The study contributes to a more nuanced understanding of language anxiety in immigrant contexts and provides evidence for the relationship between anxiety, bilingual speech and autonomic arousal.

Keywords

Language anxiety immigrant context heritage language anxiety majority language anxiety electrodermal activity skin conductance response skin conductance level bilingual speech monolingual mode bilingual mode Turkish immigrants

Introduction

In their daily lives, members of an immigrant community find themselves at various points along a situational continuum that induces a particular speech mode (Grosjean, 1982). Depending on the context, they may remain in a monolingual mode, speaking the majority language¹ with people from the mainstream community, or in a bilingual speech mode, mixing the heritage language² and majority language at home or within their ethnic community. When in bilingual speech mode, speakers borrow lexical items and practice code-switching, alternating between two languages at the word, phrase or clause levels (Valdés, 2005). Lack of practice of either the heritage or majority language in a monolingual setting may result in a decline of confidence in one’s linguistic competence. As a result, immigrants may experience anxiety when speaking their non-dominant language, whether heritage or majority, with native speakers³ of that language.

Sociolinguistic power relationships between interlocutors (see Bourdieu, 1977), for example between immigrants and so-called natives, can also heighten anxiety in regard to a bilingual’s non-dominant language use, particularly when listeners evaluate and consciously correct the bilingual’s speech (Sevinç, 2016). Social and linguistic factors can therefore invoke majority language anxiety (MLA) or heritage language anxiety (HLA) during an immigrant’s interaction with native speakers of these languages. Compared to the anxiety of foreign language learners and users, immigrant language anxiety, whether HLA or MLA, is more complex. It is also social and dynamic in nature and unique to the immigrant or minority context (Sevinç & Dewaele, 2018).

Anxiety refers to a “transitory emotional state or condition characterized by feelings of tension and apprehension and heightened autonomic nervous system activity” (Spielberger, 1972, p. 24). This is a state that can have both negative and positive effects and which both facilitates and inhibits cognitive activities such as speaking. Correspondingly, speaking is a complex operation and requires coordination of neural networks underlying cognitive, linguistic and motor processes (Arnold, MacPherson, & Smith, 2014). Hence, “the physical symptoms of speaking anxiety include sweaty palms, gastrointestinal discomfort, trembling, and increases in heart rate (Witt et al., 2006), all of which are indicative of automatic nervous system activation (Croft et al., 2004; Friedman & Thayer, 1998)” (cited in MacIntyre, MacIntyre, & Carre, 2010, p. 287).

Language anxiety is traditionally studied in language learning settings, most commonly foreign language classrooms (Horwitz, 2010). Yet the subtle effects of language anxiety can have a profound impact on individuals beyond the classroom as well (MacIntyre, 2017; Steinberg & Horwitz, 1986). Studies increasingly suggest language anxiety research should be expanded to include daily interactions of immigrant or minority communities (e.g. Garcia de Blakeley, Ford, & Casey, 2017). One such study of three generations of the Turkish immigrant community in the Netherlands demonstrated that second- and most third-generation bilinguals suffer from HLA, while first- and second-generation immigrants typically experience MLA with/around native speakers of Dutch (Sevinç & Dewaele, 2018). It also established a relationship between language background variables and HLA/MLA in the native speaker context: higher levels of immigrants’ HLA were correlated with earlier age of acquisition (AoA) of Dutch, lower proficiency in Turkish and lower frequency of daily use of Turkish. Correspondingly, lower levels of MLA were significantly linked to earlier AoA of Dutch, higher Dutch proficiency and higher frequency of daily use of Dutch.

Although the aforementioned study usefully indicates that immigrants from different generations experience different levels of HLA and MLA in different social contexts, it is exclusively based on self-reported measures. The current study, as a proof-of-concept, is based on physiological markers of immigrant language anxiety in different language modes. It examines electrodermal activity (EDA) in bilingual and monolingual modes in the use of Turkish and Dutch with/around native speakers of Turkish and Dutch. It investigates the relationship between bilinguals’ skin conductance level (SCL) and skin conductance response (SCR), self-reported HLA and MLA, and background variables (AoA, self-perceived language proficiency, frequency of daily language use). Moreover, although changes in psychophysiological stress reactivity in response to speech-related language anxiety have been reported in clinical studies (Cahana-Amitay et al., 2015), studies of public speaking anxiety (Kreibig, 2010) and foreign language anxiety (Gregersen, MacIntyre, & Meza, 2014), bilinguals’ autonomic arousal in an immigrant context and in different language modes has so far not been investigated. The purpose of the current study is thus two-fold: first, to test the presence of HLA and MLA among Turkish immigrants in the Netherlands through EDA measures; second, to provide insight into and evidence of links between language anxiety in immigrant contexts, language background variables and autonomic arousal in bilingualism.

Language anxiety and electrodermal activity

Language anxiety is defined as the worry and negative emotional reaction when learning or using a second language or foreign language (MacIntyre, 1999). Physiological indications of language anxiety include heart rate, heart rate variability and a variety of electrodermal biomarkers such as SCL. These signs have been examined in studies of public speaking among healthy adults (heart rate: MacIntyre et al., 2010; heart rate and SCL: Croft, Gonsalvez, Gander, Lechem, & Barry, 2004), foreign language learners (heart rate: Gregersen et al., 2014), and people with aphasia (heart rate and SCL: Cahana-Amitay et al., 2015).

Skin conductance is defined as a measure of EDA dependent on the activation of sweat glands (Boucsein, 2012). It consists of two components, SCL and SCR, which may rely on different neural mechanisms (Dawson, Schell, & Filion, 2007). SCL, the general tonic-level EDA, relates to the slower acting components and background characteristics of the signal (the overall level, slow rises or declinations over time). Changes in the SCL are thought to reflect general changes in autonomic arousal associated with emotional reactivity, fear and stress (see Gilissen, Koolstra, van Ijzendoorn, Bakermans-Kranenburg, & van der Veer, 2007). In contrast, SCR, the phasic component of EDA, refers to the more specific and faster changing elements of the signal in relation to stimuli presented (Braithwaite, Watson, Jones, & Rowe, 2015; Caldwell-Harris, Tong, Lung, & Poo, 2011).

In a study by Knight and Borden (1979), anticipation of public speaking led to increased SCLs. On the other hand, Geen (1989) found that during the first trial of a paired-associates learning task, subjects showed no increase in SCL, yet had a greater number of spontaneous SCRs in the presence of an observer than when alone. According to Geen, the reason participants showed increased SCRs but no change in SCL was unclear. Considering the audience to be a stimulus for evaluation apprehension during the speech, he then predicted, quoting Moore and Baron (1983), that “the SCR might reflect a basic motivational process associated with stress or threat, whereas SCL might reflect cognitive processing” (cited in Geen, 1989, p. 21). He also pointed out that these assumptions should be treated with caution since the relationship between the social presence of others and increased physiological arousal had not been systematically grounded in the field. He was also ambivalent as to whether increased SCRs were speech related since the possible relation between participants’ speech performance and EDA was not analyzed. These contrasting assumptions at the very least suggest that it is important to examine the two components of EDA, SCL and SCR, simultaneously when investigating bilinguals’ language anxiety around other people (e.g. native speakers). Furthermore, although the higher SCLs and SCRs correspond to higher autonomic arousal, in order to ensure that the increase in autonomic arousal reflects the right physiological process (e.g. language anxiety), skin conductance measurement requires additional measures such as self-reports that constrain the possible interpretations of the changes in skin conductance (Figner & Murphy, 2011).

The current study relies on electrodermal measures mainly because it focuses on arousal related both to the anticipation and performance of a distinctive speech mode (Turkish or Dutch) in a specific language mode (bilingual or monolingual). Note that again the SCL only makes up a small proportion of the overall EDA complex, and recent evidence suggests that both components (SCL and SCR) should be taken into account when analyzing EDA (see Dawson et al., 2007). This study, therefore, draws on two different measures of electrodermal activity: not only changes in SCL (an indicator of a more general level of arousal related to fear or anticipation), but also changes in SCR, which can be directly related to stimuli or performance (e.g. unknown words, code-switching, errors) during a speech (Geen, 1989).

Language anxiety in the immigrant context does not only occur in a classroom setting, but also in immigrants’ daily life during their direct or indirect interactions with native speakers of the heritage or majority language, including family members (e.g. grandparents, fathers) (Sevinç & Dewaele, 2018). It concerns various factors (linguistic, social and political) and value systems underpinned by immigrants’ identity, culture and so forth. Faced with the expectation by mainstream society and/or their ethnic community to function in both (all) languages with “nativeness” in a monolingual bias (cf. Grosjean, 2008), immigrants may experience psychological pressure in their daily interactions. This social-psychological pressure may lead to language anxiety, which is consistent with the more general concept of social facilitation (Allport, 1924). The social facilitation theory maintains that the presence of others hinders the performance of difficult tasks, possibly due to psychological pressure or stress (Geen, 1989). An audience (e.g. a native Dutch speaker in the current context) can be a stimulus for elevated apprehension and therefore cause a greater number of spontaneous SCRs associated with anxiety (Geen, 1989). The question is whether the social facilitation theory would apply when members of an immigrant or minority community speak either the heritage or majority language with/around native speakers of that language.

Language background variables (AoA, self-reported oral proficiency and frequency of language use) may also have an effect on language anxiety. Frequent use of a foreign language has been found to boost perceived competence and self-confidence and lower language anxiety (Baker & Maclntyre, 2000). Similarly, higher levels of self-perceived language proficiency are linked to lower levels of language anxiety (Santos, Cenoz, & Gorter, 2017). These perceptions are subjective (Dewaele, Petrides, & Furnham, 2008) – those who are anxious about using their languages may underestimate their proficiency, while others who are less anxious may overestimate it (MacIntyre, Noels, & Clément, 1997). There is also an ongoing debate whether AoA may affect language outcome or perception of languages. In Dewaele et al.’s (2008) study, participants who started learning a second language in early childhood reported lower levels of language anxiety. Yet the relationship between the two was not linear, which means a lower AoA may not automatically indicate a lower level of language anxiety. Investigating the link between these language background variables and EDA related to language anxiety (heritage and majority) may also shed light on the debates concerning language anxiety and language background variables.

The current study

The present study combines linguistic, social, psychological and physiological aspects to provide evidence for links between language anxiety in immigrant contexts and autonomic arousal in particular speech modes. The study asks whether bilinguals’ physiological reactivity can confirm the presence of heritage language anxiety and majority language anxiety among Turkish immigrants in the Netherlands. Specifically, the two following questions are addressed:

What are the differences in autonomic arousal (electrodermal activity – skin conductance level and skin conductance response) between Turkish–Dutch bilinguals across three generations when they use the heritage language (Turkish) versus the majority language (Dutch) in monolingual speech mode with/around a native speaker?

What is the relationship between autonomic arousal, self-reports on language anxiety and language background variables (age of acquisition, self-reported oral proficiency and frequency of language use)?

Based on participants’ previous self-reports on heritage language anxiety and majority language anxiety and language background variables (see Table 1), it is predicted that third-generation bilinguals, who acquired Dutch earlier, who have higher oral Dutch proficiency and who use Dutch more frequently than first- and second-generation bilinguals, will have higher SCL and SCR amplitude during Turkish monolingual mode than Dutch monolingual mode. Their autonomic arousal will also be greater than first- and second-generation bilinguals’ autonomic arousal during Turkish monolingual mode.

Table 1.

Participants’ demographic information and language background.

	1^st gen.			2^nd gen.			3^rd gen.
	(n=6)			(n=8)			(n=16)
	Mean	SD	Range	Mean	SD	Range	Mean	SD	Range
Age (Years)	41	4.0	33–43	26	1.6	24–28	14	2.2	12–19
Gender
Female (n=21)		6			8			7
Male (n=9)		−			−			9
	Mean	SD		Mean	SD		Mean	SD
Age of Dutch acquisition	24.0	5.3		3.8	0.5		0.4	1.0
Age of Turkish acquisition	0.0	0.0		0.0	0.0		0.0	0.0
Oral proficiency in Dutch	3.3	0.5		4.1	0.4		4.7	0.5
Oral proficiency in Turkish	4.8	0.4		3.7	1.0		2.9	0.9
Frequency of Dutch use	4.0	0.9		4.3	0.7		4.8	0.4
Frequency of Turkish use	4.2	0.7		3.3	1.2		2.6	0.5
MLA with Dutch people	1.5	0.5		2.0	0.9		1.0	0.2
HLA with Turks in Turkey	1.0	0.0		1.5	0.5		3.5	1.4

Range of oral proficiency: (1) None; (2) Poor; (3) Fair; (4) Good; (5) Excellent.

Range of language use per day: (1) Never; (2) Rarely; (3) Sometimes; (4) Frequently; (5) All the time.

Range of HLA and MLA: (1) Not at all; (2) A little; (3) Quite anxious; (4) Very anxious; (5) Extremely anxious.

HLA: heritage language anxiety; MLA: majority language anxiety.

Method

Participants

Thirty individuals from the Turkish immigrant community in the Netherlands (21 female, nine male) participated in this study; six were first-generation bilinguals, eight were second-generation, and 16 were third-generation (Table 1). One of the first-generation participants came to the Netherlands through labor migration and five of them came through marriage migration after marrying their second-generation Turkish spouse. The term “second generation” in this study refers to Turkish people who were born in the Netherlands or arrived there before the age of five and started school there. “Third-generation” participants were bilinguals who had one second-generation Turkish–Dutch bilingual parent and one parent born in Turkey who came to the Netherlands through marriage migration.

First-generation immigrants completed their education in Turkey (two were university graduates, one held a high school diploma and two completed elementary school). Second-generation bilinguals varied in their educational levels from high school to higher education in the Netherlands, while third-generation bilinguals were enrolled either in elementary school or high school in the Netherlands. Table 1 also illustrates 30 participants’ language background based on a five-point Likert scale questionnaire they filled out one month prior to the experiment.

All participants acquired Turkish from birth, but third-generation participants were exposed to Dutch earlier than first- and second-generation participants. Self-reported oral language proficiency and frequency of language use of participants differed across the three generations. In comparison to first- and second-generation participants, third-generation participants’ oral proficiency and daily language use ranked lowest in Turkish and highest in Dutch. Third-generation bilinguals reported the highest language anxiety when speaking Turkish with/around Turks, while they reported almost no anxiety when speaking Dutch in any conditions. First- and second-generation participants, on the other hand, reported anxiety when speaking Dutch, while second-generation bilinguals reported also experiencing anxiety when speaking Turkish.

Equipment

A Biopac system consisting of the base module (MP150WS) and the modules for skin conductance (GSR 100C) and cardiovascular activity (PPG100C) was used. AcqKnowledge 3.9 software used for real-time monitoring, recording the data to a hard drive, and for data filtering and analysis. A laptop with a 24” LCD monitor was used to administer the experimental task. One video camera and two audio recorders were used.

Disposable electrodes were chosen. Unlike reusable ones, they were pre-gelled and did not require any cleaning or disinfection after use. Given that external factors such as temperature and humidity affect skin conductance measurements (Boucsein, 2012), a digital hygrometer/thermometer was used to measure the temperature and humidity in the room. Data analyses were carried out in AcqKnowledge 4.1.

Experimental task

The experiment consisted of a video-retelling task. Forty-two video clips⁴ were shown. They varied in duration from nine to 80 seconds. Fourteen video clips were used during the two free-language modes. Twenty-eight video clips were played during the Dutch and Turkish monolingual modes. To limit the duration of the experiment, fewer video clips were presented in the free modes than in the monolingual modes, the main focus of the study. Importantly, the video clips contained no emotional components. This was to ensure that the language mode as the only factor influencing emotional arousal. The clips portrayed daily activities (washing something, eating an apple, drinking water, etc.). They were designed to elicit the production of specific structures such as case markers, word order variations, plural suffixes, postpositions, etc., as well as a varied vocabulary.

The experiment was carried out in six phases: baseline1, free-mode1, Turkish monolingual mode, Dutch monolingual mode, free-mode2 and baseline2. The total duration of the video description tasks was 25–30 minutes. To establish a baseline for arousal, the baseline mode and the free-language modes preceded the monolingual modes. To prevent any order bias, the experiment was performed in two versions in which only the order of the Turkish and Dutch monolingual modes differed. After completing baseline1 and free-mode1, 14 participants started with the monolingual mode in their non-dominant, least comfortable language, while the other 14 did the opposite.

Two second-generation participants who selected both Turkish and Dutch as their dominant language started the experiment in Turkish monolingual mode (Table 2). Sixteen of the participants, therefore, performed the Turkish monolingual mode before they performed the Dutch monolingual mode. Table 2 shows the distribution of participants across generations, languages and orders.

Table 2.

Experimental order and participants; self-reports on the dominant language.

	1^st gen.			2^nd gen.			3^rd gen.
	(n=6)			(n=8)			(n=16)
Dominant language	Dutch	Turkish	Both	Dutch	Turkish	Both	Dutch	Turkish	Both
	0	6	0	3	3	2	16	0	0
Experimental order	NLM first	TRM first		NLM first		TRM first	NLM first		TRM first
	3	3		3		5	8		8

NLM: Dutch monolingual mode; TRM: Turkish monolingual mode.

Procedure

To create a natural setting as opposed to a laboratory, the experiment was conducted in the author’s house. Participants were tested individually. The two researchers carrying out the experiment first introduced themselves and the equipment and then explained that one of them came from Turkey and knew no Dutch and the other one came from the Netherlands and had no knowledge of Turkish.

Prior to the experiment

Overall instructions were given in both Turkish and Dutch. Participants were then asked which language they would prefer. To reduce anxiety related to the experimental setting, participants were informed they would not be limited by time nor scored on their language proficiency. They were instructed to describe the video clips while watching them and informed that they were free to pause the clips at any time. This was intended to reduce the experimental time-dependent stress. The video camera was positioned about four meters from the participants to minimize discomfort.

To prevent a poor skin conductance signal, participants washed and dried their hands. Any jewelry near the electrode sites was removed. The two electrodes for skin conductance were placed on the distal (first) phalanges of the index and middle finger of the non-dominant hand so the participants could write or use the keyboard if necessary. Participants were asked to fill out the consent form and provide some background information about themselves and their language. This gave the electrode gel enough time to soak into the skin properly and detect the conductance, resulting in a more stable electrical connection.

During the experiment

Participants’ speech was audio recorded, and their faces and hands were video recorded to allow for judgments of speech fluency. During the treatment of the raw data following the experiment, the video recordings also helped exclude periods contaminated with movement artifacts. One of the researchers monitored the physiological data recording, while the other was controlling the monolingual mode in her/his native language. During the baseline and free modes, the experimenters controlling the experiment were positioned at a distance of approximately three meters from the subjects. Throughout the physiological data recording, any events that could inadvertently influence results (e.g. sudden movements or coughs) were marked manually by one of the experimenters using the event-marking feature of AcqKnowledge.

There was a one- to two-minute interval after each phase, allowing the researchers to check on the preceding phase and the participant to rest for at least 30 seconds. The experiment contained the following six phases.

Baseline1: The baseline mode establishes the individual resting level of arousal during non-activity and whether participants are likely to be hyper- or hypo-responders independent of any effects of psychological manipulation (Braithwaite et al., 2015). Before the experiment, participants were asked to take deep breaths in order to check whether all the signals were working. During the baseline mode, they were asked to close their eyes and relax for approximately two minutes.

Free-mode1: Seven video clips were shown, and participants were asked to describe them in whatever language, or languages, they felt comfortable with. The purpose of this phase was to familiarize participants with the experiment under non-stressful circumstances and thereby minimize the effect anxiety would have on subsequent phases of the experiment.

Monolingual-mode1: 28 video clips were viewed and described in monolingual mode by the participants, either in Turkish with the Turkish researcher or in Dutch with the Dutch researcher, depending on informants’ self-reports of their dominant language (Table 2).

Monolingual-mode2: Participants described the same 28 video clips as in the previous exercise but this time in the other language.

Free-Mode2: Seven video clips were played, and participants were instructed to describe the videos in whatever language they felt comfortable with or to switch between languages if needed.

Baseline2: Participants were asked to close their eyes and take deep breaths.

Free-mode2 and baseline2 allowed for any changes in EDA to be detected after the monolingual modes. They also allowed for a continued high signal quality to be established at the end of the experiment.

After the experiment

After participants had completed all six phases of the experiment, the electrodes were removed. Participants were asked to report the language anxiety levels that they experienced during the two monolingual modes on a 5-point Likert scale ranging from none (1) to extreme (5).

Analysis

Methodological considerations and signal analysis

The acquisition parameters of the Biopac hardware and GSR100C amplifier settings were set as follows: amplification was set to 5 μS/V, the low-pass filter to 1Hz and no hardware high-pass filters were activated (i.e. the switches were set to DC, see Figner & Murphy, 2011). Data were initially sampled at 1kHz, and the EDA signals were recorded with a gain of 5 uS/V. To prevent potential software lag during data collection, filtering and smoothing processes were carried out after the raw signal had been saved.

The SCL, the background tonic EDA, can differ markedly between individuals. It is therefore unclear whether any given overall SCL measured can be interpreted as “high” or “low” for that individual. In addition, simply averaging SCL across the whole signal is inadequate as a measure of SCL because it will contain SCRs, which artificially elevate the measure (Boucsein, 2012). Therefore, EDA data require certain normalization procedures prior to analysis. It is most of all necessary to subtract the amplitudes of SCRs from the tonic signal to establish a truer representation of background SCL, or to take measurements from periods outside of SCRs in the signal (Boucsein, 2012). Further, it must be possible to subtract experimental phases from each other such that the resultant SCL measure is a relative difference across manipulations within an individual (Boucsein, 2012). The subtraction procedure acts as a form of normalization for the participant’s EDA data.

After the subtraction procedure, SCRs should be detected – namely, the amplitude of event-related SCR peaks (event amplitude)⁵ which are “attributed to a specific elicited stimuli” (Braithwaite et al., 2015, p. 4). Typically, an event-related SCR is defined as an SCR whose latency period between stimulus onset and the first significant deviation in the signal is between 1 and 3 seconds (Figner & Murphy, 2011). Deflections in the signal that occur before this period are typically defined as non-specific SCRs and are not viewed as coming directly from experimental manipulations. Based on these considerations, the following data treatment steps were performed:

First, the raw data was cleaned. All data sets containing raw skin conductance signals were explored using AcqKnowledge 4.1 for measures of skew, kurtosis and heterogeneity of variance before a formal analysis was carried out. No transformation was required since the SCLs and SCRs were normally distributed. Researchers viewed the video and audio files recorded during the experiments and monitored the event markers they had previously noted on the raw data (see Appendix 1: Figures 1.1, 1.2 and 1.3 for samples of raw data). Any responses directly related to stimuli (i.e. participants’ speech performance, e.g. hesitations, pauses, code-switching, using fillers for unknown words) were kept, whereas other SCRs unrelated to speech (e.g. sudden movements, sneezes or coughs) were removed from the raw data. Second, a high-pass filter was applied in AcqKnowledge. The current preference for the SCR threshold was about 0.03μS (with a range of 0.01μS–0.03μS). Setting the threshold to zero and the rejection rate to 10% closely approximates the SCR detection algorithm outlined in Kim, Bang, and Kim (2004). Mean values of the SCR amplitudes, duration and the number of SCR occurrences (frequency) in a 50s signal segment were extracted from the EDA as features. Detected SCRs with an amplitude smaller than 10% of the maximum SCR amplitude in this segment were excluded (see Kim et al., 2004).

Following Boucsein (2012), the amplitudes of SCRs were subtracted from the tonic signal (SCL) in order to establish a more reliable representation of background SCL. SCLs and SCRs were therefore set to be analyzed separately. The mean measurements of the SCL were taken from periods outside of SCRs in the signal for each experimental phase. The measures from the six phases were averaged separately to produce reference values. Then, locating SCRs using the “locate SCRs” function⁶ of AcqKnowledge 4.1, all peaks were verified in the software by the two experimenters.

To facilitate an individual-difference analysis and improve the validity and reliability of the data set, changes in the mean scores of participants’ SCLs for each phase relative to the mean baseline measured during baseline1 were established individually. For the purpose of this study, the main comparison was made between the mean scores in the Dutch and Turkish monolingual modes. The mean SCL and SCR amplitude in the Dutch and Turkish monolingual modes were subtracted from each other. After this subtraction, greater SCL and SCR values were considered to signify greater tonic and phasic levels of sweat secretion (Dawson et al., 2007). This subtraction procedure constitutes a form of normalization of the participants’ EDA data. Given that speech-related sympathetic arousal may decrease with age as individuals develop greater proficiency with speech tasks (Arnold et al., 2014) and that the age difference is large across generations in the current data, the subtraction procedure also helped eliminate the effects of certain gross influences due to age differences. Although younger participants may have higher SCLs than older ones in general, comparing each participant’s mean EDA measurements across different language modes eliminated age-related differences in EDA and isolated autonomic arousal related to a specific language mode.

Statistics

The data were analyzed to assess the main research questions related to generation effects in autonomic measures during the Dutch and Turkish monolingual modes and the correlations between autonomic arousal and the questionnaire data. Prior to the analyses linked to the research questions, any effects of gender and experimental order (i.e. whether participants completed the Dutch or Turkish monolingual mode first) on SCL and SCR measurements for each of the six phases were tested using independent sample t-tests. There was no significant effect of either variable.

Two sets of analyses were performed to address the research questions. First, based on descriptive analysis of the EDA measurements (means and standard deviations across three generations and six experimental phases), a series of ANOVAs were run to evaluate the effects of generation on skin conductance activity.

Second, assumptions of normal distribution of the questionnaire data were checked for the language background variables (self-reported oral language proficiency and frequency of daily use of the languages) and levels of HLA and MLA (Kolmogorov-Smirnov Z-values varied between 0.26 and 0.44 – all significant at p < .0001). Correlations between skin conductance activity and the questionnaire data (i.e. language background variables and self-reports) were examined using Spearman’s rho as nonparametric equivalents to Pearson’s r. The Holm’s sequential Bonferroni method correction was used to control for the increased risk of Type I error associated with multiple comparisons (see Holm, 1979).

Results

First, findings on the EDA measures (i.e. SCL and SCR measures) across generations and speech phases are provided. Then, the correlation analyses are presented to explore a possible link between the SCL and SCR measures, self-reported heritage language anxiety and majority language anxiety, and language background variables (i.e. AoA, self-reported oral language proficiency and daily use of Turkish and Dutch).

EDA (SCL/SCR) across generations and different experimental phases

Skin conductance level

Generation effect on SCL across four speech modes

Figure 1 shows SCL values across the four speech modes and three generations. A mixed design ANOVA revealed a statistically significant main effect of generation, F(2,27) = 6.98, p < 01, $η_{p}^{2} =$ .34, and a significant interaction between speech mode and generation, F(6,81) = 7.17, p < .001, $η_{p}^{2} =$ .35. Bonferroni post-hoc tests revealed that the mean SCL for first-generation participants was significantly lower in the four speech modes (p < .01, p < .001, p < .05, p < .01, respectively) compared to third-generation participants, while it was significantly lower only in the free-mode2 compared to second-generation participants (p < .05) (Figure 1). There were no significant differences between the second- and third-generation groups in any of the speech modes. This is likely the result of age differences across generations (see Arnold et al., 2014). To probe that relationship, a Pearson correlation analysis of SCL and age was performed (Table 3).

Table 3.

Relationship between age and mean skin conductance level (SCL) in six phases.

Age	SCL-BLM1	SCL-FM1	SCL-TRM	SCL-NLM	SCL-FM2	SCL-BLM2
ρ	−.54^† **	−.54^† **	−.61^† ***	−.38^† *	−.52^† **	−.51^† **
p	0.001	0.002	0.000	0.036	0.003	0.003
n	30	30	30	30	30	30

†

Significant after Holm’s correction for multiple comparisons.

p < .05; ^**p < .01; ^***p < .001(all two-tailed tests).

BLM1: Baseline1; FM1: Free-mode1; NLM: Dutch monolingual mode;

TRM: Turkish monolingual mode; FM2: Free-mode2; BLM2: Baseline2.

Table 3 presents the results of a Pearson correlation analysis of SCL and age. The analysis revealed a significant negative correlation between the SCL measures and age in all six phases. As anticipated, younger participants (third-generation bilinguals) in general had higher SCL measures than older ones, yet the subtraction results indicated that third-generation bilinguals also had increased autonomic arousal in the Turkish monolingual mode relative to the Dutch mode.

Mean SCL change in speech modes relative to baseline

Figure 2 shows the change of mean SCL in each of the four speech modes across generations, relative to the mean SCL baseline. A mixed design ANOVA revealed no main effect of generation on mean SCL change in the four speech modes relative to the baseline, but a significant interaction between mode and generation, F(6,81) = 7.25, p < .001, $η_{p}^{2} =$ .35. Bonferroni post-hoc tests revealed that the mean SCL change in the Turkish monolingual mode for third-generation participants was significantly higher than for first-generation participants, p < .05 (Figure 2).

As seen in Figure 2, second- and third-generation bilinguals also experienced high autonomic arousal during the free modes (particularly in free-mode2). The reason for the anxiety in a free mode when participants could mix the two languages freely was not systematically examined in this study. Possible reasons will be addressed in the conclusion.

Mean SCL change across monolingual modes

Figure 3 illustrates the mean SCL difference between the Turkish and Dutch monolingual modes. One-way ANOVAs on each language mode revealed statistically significant main effects of generation for SCL related to heritage and majority language anxiety – for the mean SCL difference between Turkish monolingual mode and Dutch monolingual mode – F(2,27) = 19.3, p < .001. A Bonferroni post-hoc test revealed that the mean difference between the Turkish and Dutch monolingual modes for third-generation bilinguals was significantly greater than the mean difference for first- and second-generation bilinguals (p < .01 and p < .001, respectively; see Figure 3). These findings indicate an increased SCL during the Turkish monolingual mode for third-generation bilinguals. Conversely, compared to the Turkish monolingual mode, the mean SCL difference was higher in the Dutch monolingual mode for first- and second-generation bilinguals (p < .01 and p < .001, respectively) than third-generation bilinguals.

Skin conductance response

Generation effect on SCR across the four speech modes

Figure 4 shows the mean SCR event amplitude in each of the four speech modes across generations. A mixed design ANOVA revealed a statistically significant main effect of generation for SCR amplitude during the four speech modes, F(2,27) = 6.83, p < .01, $η_{p}^{2} =$ .34; for modes, F(3,81) = 5.48, p < .01, $η_{p}^{2} =$ .17; and a significant interaction between mode and generation, F(6,81) = 6.85, p < .001, $η_{p}^{2} =$ .34. As in the case of the overall SCL, Bonferroni post-hoc tests revealed no significant differences between the second- and third-generation groups. The mean SCR event amplitude of first-generation participants was significantly lower than those of second-generation bilinguals (in Turkish monolingual mode p < .05; in free-mode2 p < .01) and third-generation bilinguals (in free-mode1 p < .01; in Turkish monolingual mode p < .001; in free-mode2 p < .01) (see Figure 4). These results can also be linked to age.

Note that again second- and third-generation bilinguals had high levels of mean SCR amplitude during the free modes, particularly in free-mode2. For third-generation bilinguals, the mean SCR amplitude in free-mode1 and free-mode2 was higher than in the Dutch monolingual mode, while for second-generation bilinguals, it was higher in free-mode2 than in the Turkish monolingual mode.

Change in SCR event amplitude across monolingual modes

Figure 5 shows the mean SCR amplitude difference between monolingual modes. One-way ANOVAs revealed statistically significant main effects of generation for SCR event amplitude difference between Turkish and Dutch monolingual modes, F(2,27) = 28.0, p < .001.

A Bonferroni post-hoc test revealed that the mean SCR event amplitude difference between Turkish and Dutch monolingual modes for third-generation bilinguals was significantly greater than the mean difference for first- and second-generation bilinguals, p < .001 and p < .001, respectively (Figure 5).

EDA (SCL/SCR) versus self-reported anxiety, and questionnaire data

Table 4 presents the participants’ self-rated level of anxiety experienced in Turkish and Dutch monolingual modes. Third-generation bilinguals reported the highest language anxiety in the Turkish monolingual mode (HLA), while they reported no anxiety in the Dutch monolingual mode (MLA). First- and second-generation participants scored higher on MLA than HLA.

Table 4.

Participants’ self-reports on heritage language anxiety and majority language anxiety after the experiment.

	1^st gen.		2^nd gen.		3^rd gen.
	(n=6)		(n=7)		(n=16)
	M	SD	M	SD	M	SD
Self-rated heritage language anxiety	1.0	0.0	1.3	0.5	3.8	1.2
Self-rated majority language anxiety	2.5	0.8	2.7	1.2	1.1	0.2

Anxiety scale: (1) Not at all; (2) A little; (3) Quite; (4) Very anxious; (5) Extremely anxious.

A Pearson correlation analysis of SCL and SCR measures and the self-reported anxiety levels in the monolingual modes revealed significant correlations (Table 5). As signs of HLA and MLA, levels of the mean difference in the EDA measurements between Dutch and Turkish monolingual modes were found to be higher for participants who reported higher levels of language anxiety during these monolingual modes.

Table 5.

Correlations between mean EDA difference between monolingual modes and self-reports on heritage language anxiety (HLA) and majority language anxiety (MLA).

EDA in NLM minus EDA in TRM (MLA)	SCL	SCR
Self-rated majority language anxiety
ρ	.83^† *	.77^† *
p	.000	.000
n	30	30
EDA in TRM minus EDA in NLM (HLA)	SCL	SCR
Self-rated heritage language anxiety
ρ	.77^† *	.79^† *
p	.000	.000
n	30	30

Anxiety scale: (1) not at all; (2) a little; (3) quite; (4) very anxious; (5) extremely anxious.

EDA: electrodermal activity; TRM: Turkish monolingual mode; NLM: Dutch monolingual mode; SCL: skin conductance level; SCR: skin conductance response.

†

Significant after Holm’s correction for multiple comparisons.

p < .001 (all two-tailed tests).

A further Spearman correlation analysis of the SCL and SCR in monolingual modes and background variables (Table 6) revealed that SCLs and SCRs in monolingual modes were significantly correlated with AoA of Dutch. Lower levels of autonomic arousal in the Dutch monolingual mode and higher levels of autonomic arousal in the Turkish monolingual mode were significantly linked to earlier AoA of Dutch. Dutch oral proficiency levels were not correlated with any of the EDA measures, while frequency of daily use of Dutch was negatively correlated only with the SCL in the Dutch monolingual mode. The SCL was found to be lower for participants who reported a higher frequency of Dutch use.

Table 6.

Relationship between mean EDA difference between monolingual modes and language background variables.

EDA in NLM minus TRM (MLA)	SCL	SCR
Language background variables
Age of Dutch acquisition
ρ	.74^† ***	.81^† ***
p	.000	.000
n	30	30
Oral proficiency in Dutch
ρ	−.36	−.43
p	.058	.058
n	30	30
Frequency of Dutch use
ρ	−.48^† *	−.34
p	.037	.137
n	30	30

EDA in TRM minus NLM (HLA)	SCL	SCR
Language background variables
Age of Dutch acquisition
ρ	−.74^† ***	−.81^† ***
p	.000	.000
n	30	30
Oral proficiency in Turkish
ρ	−.52^† **	−.64^† ***
p	.003	.000
n	30	30
Frequency of Turkish use
ρ	−.44^† *	−.47^† *
p	.017	.017
n	30	30

EDA: electrodermal activity; TRM: Turkish monolingual mode; NLM: Dutch monolingual mode; SCL: skin conductance level; SCR: skin conductance response; MLA: majority language anxiety.

†

Significant after Holm’s correction for multiple comparisons.

p < .05; ^**p < .01; ^***p < .001 (all two-tailed tests).

Higher levels of SCLs and SCRs in the Turkish monolingual mode were negatively correlated with participants’ AoA of Dutch, Turkish oral proficiency levels and frequency of daily use of Turkish. Results on SCL and SCR amplitude in Turkish monolingual mode were high for participants who reported earlier AoA of Dutch, lower oral proficiency in Turkish and lower frequency of daily use of Turkish.

Finally, the link between age and autonomic arousal during Turkish and Dutch monolingual modes needs to be clarified. Although younger participants (third-generation bilinguals) had higher SCLs and SCRs than older ones in general, findings based on the subtraction process of the two monolingual modes indicated negative correlations between age and EDA measures in the Turkish monolingual mode. Younger bilinguals had increased autonomic arousal in the Turkish monolingual mode more than in the Dutch monolingual mode (Table 7). Correlations were positive for EDA measures in the Dutch monolingual mode, which indicates that older participants had a higher autonomic arousal in the Dutch monolingual mode than in the Turkish monolingual mode.

Table 7.

Relationship between age and mean EDA differences between monolingual modes.

Age	SCL_TRM minus_NLM	SCL_NLM minus_TRM	SCR_TRM minus_NLM	SCR_NLM minus_TRM
ρ	−.60^† *	.60^† *	−.78^† *	.77^† *
p	0.000	0.000	0.000	0.000
n	30	30	30	30

EDA: electrodermal activity; TRM: Turkish monolingual mode; NLM: Dutch monolingual mode; SCL: skin conductance level; SCR: skin conductance response.

†

Significant after Holm’s correction for multiple comparisons.

p < .001 (all two-tailed tests).

Discussion and concluding remarks

The first aim of this study was to investigate whether differences in the level of EDA (i.e. SCL/SCR) as an indicator of HLA and MLA were present among Turkish immigrants across three generations when they used Turkish and Dutch in monolingual modes in the presence of native speakers of either Turkish or Dutch.

The results revealed statistically significant effects of generation on changes in EDA (SCL and SCR amplitudes) between the two monolingual modes. During the Turkish monolingual mode, third-generation bilinguals experienced the greatest EDA. In comparison to their arousal level during the speech task in the Dutch monolingual mode, third-generation bilinguals demonstrated greater change in arousal levels during the speech task in the Turkish monolingual mode than first- and second-generation bilinguals, suggesting a greater degree of anxiety for third-generation bilinguals when speaking the heritage language. Conversely, compared to the Turkish monolingual mode, the mean difference in EDA was higher in the Dutch monolingual mode for the first- and second-generation groups, suggesting greater anxiety levels when speaking the majority language.

Furthermore, second- and third-generation bilinguals also had high levels of EDA during the free modes, particularly in free-mode2. For third-generation bilinguals, autonomic arousal in free modes (1 and 2) was higher than in the Dutch monolingual mode, while for second-generation bilinguals it was higher in free-mode2 than in the Turkish monolingual mode. The reason for this may be that second- and third-generation bilinguals mixed the two languages during free modes more frequently than first-generation participants did, who mostly used Turkish, their dominant language, without switching to Dutch. Note that although the researchers kept their distance from participants during the free modes of the experiment, they were still in the same room with them. Code-switching or mixing the two languages in the presence of the researchers who did not have an immigrant or Turkish–Dutch bilingual background might also have triggered these participants’ linguistic anxiety during free modes.

The current data demonstrated that second- and third-generation bilinguals experienced higher autonomic arousal in bilingual mode in the presence of native speakers than in the monolingual mode of the language with which they feel comfortable – Turkish for the second generation, Dutch for the third generation. However, this interpretation must be treated with caution since the conditions of the free mode phases were different from those of the monolingual modes and self-reports on participants’ language anxiety in bilingual modes were lacking in the current study. Still, this calls for further research on immigrant, or minority, community members’ feelings when mixing the two languages in the presence of people from the mainstream community. As another suggestion for future study, an experimenter with an immigrant background could control the experiment in free-mode2 if that would be preferred by participants. It would also be useful to ask participants to report their levels of anxiety during free modes and to interview them about the reasons for their anxiety when code-switching or mixing the two languages.

The study also aimed to examine the relationship between physiology (EDA), self-reports on language anxiety and language background variables (AoA, self-reported oral proficiency and frequency of language use). Importantly, the findings on physiological measures of anxiety were consistent with participants’ self-reports on the levels of HLA and MLA during Turkish and Dutch monolingual modes. These results also support previous findings that predominantly third-generation bilinguals experienced HLA with/around native speakers of Turkish (the heritage language), whereas first- and second-generation bilinguals typically experienced MLA with/around native speakers of Dutch (the majority language) (Sevinç & Dewaele, 2018).

A Spearman correlation analysis revealed that the levels of SCL and SCR during the monolingual modes were significantly correlated with the AoA of Dutch. Participants who acquired Dutch earlier had lower SCLs and SCRs during the Dutch monolingual mode than the Turkish monolingual mode. Moreover, results on SCL and SCR amplitude during the Turkish monolingual mode were higher for participants who reported lower oral proficiency in Turkish and lower frequency of daily use of Turkish. Dutch oral proficiency levels, on the other hand, were not correlated with any of the EDA measures, while frequency of Dutch use was negatively correlated only with SCL in the Dutch monolingual mode. SCL in the Dutch monolingual mode was found to be lower for participants who reported higher frequencies of daily use of Dutch.

These findings suggest that MLA is related more to social and psychological factors than to lack of language knowledge or proficiency (Sevinç & Dewaele, 2018). Yet it bears restating that these conclusions must be treated carefully, since they are based on self-rated proficiency levels. The present study assessed bilinguals’ SCLs and SCRs during a speaking task, but did not assess the association between these physiological measures and instances of language contact and change phenomena (i.e. vocabulary and sentence structure). Evaluation of participants’ actual speech during the speech modes and a comparison between their performance and EDA measurements could better determine the link between language proficiency, anxiety and EDA. A comparison could also be made with a control group that includes Turks living in Turkey and Dutch people when they complete the experiment in their mother tongue.

This is the first study to compare EDA in terms of language anxiety in bilinguals. It remains a proof-of-concept study that requires further investigation. Still, it provides evidence that language anxiety occurs in the immigrant or minority context. It shows that members of an immigrant community face psychological challenges related to the use of heritage and majority languages in various social contexts and language modes. Although bilingualism is often an advantage, it may also trigger complex emotions in which negative feelings (e.g. shame, disappointment, frustration, stress and anxiety) predominate, particularly in the immigrant context (Sevinç & Dewaele, 2018). It is therefore vital to investigate the social and psychological outcomes of bilingualism and its challenges among immigrants.

The study also draws attention to the potential effect of the presence of other people to heighten distraction and arousal, a phenomenon discussed by social facilitation theorists. Participants across generations showed an increased arousal with/around a native speaker of their less dominant language. However, the interpretation of this must be treated with caution, since the set up of the current experiment was not primarily designed to determine whether the social facilitation theory was useful in understanding immigrant language anxiety. Further studies are necessary to compare bilinguals’ arousal in different language modes when they complete the task in the presence of native speakers and when they are alone (see Geen, 1989).

Although the current study offers a new perspective, at present the best means for revealing the link between speech-related anxiety, autonomic arousal and language background variables still remains unclear. The possible relationship between physiological behaviors and specific occurrences of bilingual and monolingual speech among immigrants is an issue that awaits future study. It requires interdisciplinary research that combines linguistic factors (vocabulary and sentence structure), social factors (e.g. social exclusion and immigrant experience), psychological outcomes (language anxiety) and physiological behaviors (autonomic arousal). Future methods of investigation should therefore include physiological measures, interviews, speech analysis and an experimental design capable of assessing the effect of social facilitation and language anxiety on immigrants.

Footnotes

Appendix 1: samples of the raw EDA data

Acknowledgements

The author thanks Dr. Sandra van Dulmen and Mara Van Oost at the Netherlands Institute for Health Services Research (Nivel) in Utrecht for lending a Biopac system for the physiological data collection, Dr. Robert Levenson and his team at the Berkeley Psychophysiology Laboratory for the appreciated advice on data analysis, and Dr. Özlem Ayduk and Jyoti Jhita for welcoming her at the Relationships and Social Cognition Laboratory (RASCL) at UC Berkeley. She further thanks Dr. Marianne Gullberg and the anonymous reviewers for their insightful comments and constructive suggestions on an earlier version of this article.

The data used in this study were designed for the author’s PhD project and the first version of this manuscript has been included in the author’s PhD thesis, which is currently in the process of evaluation.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was financially supported by the Research Council of Norway through its Centres of Excellence funding scheme, project number 223265.

Notes

Author biography

Yeşim Sevinç is a doctoral student at University of Oslo, Center for Multilingualism in Society Across the Lifespan. She received her BA degree in TESL at Istanbul University and her first MA degree in Linguistics at Kafkas University. She obtained her second MA degree in Linguistics at Radboud University Nijmegen. In her first MA thesis, she studied individual differences between Polish and Turkish ESL learners. With her second MA thesis on the Turkish immigrant community in the Netherlands, she won the third best MA thesis prize in Belgium and the Netherlands awarded by Anela/VIOT in 2012. Her research interests include linguistic, social, psychological and physiological aspects of multilingualism with a particular focus on the immigrant context.

References

Allport

F. H.

(1924). Social psychology. Boston, MA: Houghton-Mifflin.

Arnold

H. S.

MacPherson

M. K.

Smith

(2014). Autonomic correlates of speech versus nonspeech tasks in children and adults. Journal of Speech, Language, and Hearing Research, 57, 1296–1307.

Baker

S. C.

MacIntyre

P. D.

(2000). The role of gender and immersion in communication and second language orientations. Language Learning, 50(2), 311–341.

Boucsein

(2012). Electrodermal activity (2nd Ed.). New York: Springer.

Bourdieu

(1977). Cultural reproduction and social reproduction. In Karabel

Halsey

A. H.

(Eds.), Power and ideology in education (pp. 487–511). New York: Oxford University Press.

Braithwaite

Watson

D. G.

Jones

Rowe

(2015). A guide for analyzing electrodermal activity (EDA) & skin conductance responses (SCRs) for psychological experiments. Technical Report: Selective Attention & Awareness Laboratory (SAAL) Behavioural Brain Sciences Centre, University of Brimingham, UK.

Cahana-Amitay

Oveis

A. C.

Sayers

J. T.

Pineles

S. L.

Spiro

Albert

M. L.

(2015). Biomarkers of “linguistic anxiety” in aphasia: a proof-of-concept case study. Clinical Linguistics & Phonetics, 29, 401–413.

Caldwell-Harris

C. L.

Tong

Lung

Poo

(2011). Physiological reactivity to emotional phrases in Mandarin–English bilinguals. International Journal of Bilingualism, 15(3), 329–352.

Croft

R. J.

Gonsalvez

C. J.

Gander

Lechem

Barry

R. J.

(2004). Differential relations between heart rate and skin conductance, and public speaking anxiety. Journal of Behavior Therapy and Experimental Psychiatry, 35, 259–271.

10.

Dawson

M. E.

Schell

A. M.

Filion

D. L.

(2007). The electrodermal system. In Cacioppo

J. T.

Tassinary

L. G.

Berntson

G. G.

(Eds.), Handbook of psychophysiology (3rd ed.) (pp. 159–181). New York: Cambridge University Press.

11.

Dewaele

J.-M.

Petrides

K. V.

Furnham

(2008). Effects of trait emotional intelligence and sociobiographical variables on communicative anxiety and foreign language anxiety among adult multilinguals: a review and empirical investigation. Language Learning, 58, 911–960.

12.

Figner

Murphy

R. O.

(2011). Using skin conductance in judgment and decision making research. In Schulte-Mecklenbeck

Kuehberger

Ranyard

(Eds.), A handbook of process tracing methods for decision research (pp. 163–184). New York: Psychology Press.

13.

Friedman

B. H.

Thayer

J. F.

(1998). Anxiety and automatic flexibility: A cardiovascular approach. Biological Psychology, 47, 243–263.

14.

Garcia

Blakeley

Ford

Casey

(2017). Second language anxiety among Latino American immigrants in Australia. International Journal of Bilingual Education and Bilingualism 20(7), 759–772.

15.

Geen

R. G.

(1989). Alternative conceptions of social facilitation. In Paulus

P. B.

(Ed.), Psychology of group influence (2nd ed.) (pp. 15–51). Hillsdale, NJ: Erlbaum.

16.

Gilissen

Koolstra

C. M.

van Ijzendoorn

M. H.

Bakermans-Kranenburg

M. J.

van der Veer

(2007). Physiological reactions of preschoolers to fear-inducing film clips: effects of temperamental fearfulness and quality of the parent-child relationship. Developmental Psychobiology, 49(2), 187–195.

17.

Gregersen

Macintyre

P. D.

Meza

M. D.

(2014). The motion of emotion: idiodynamic case studies of learners’ foreign language anxiety. The Modern Language Journal, 98, 574–588.

18.

Grosjean

(1982). Life with two languages: an introduction to bilingualism. Cambridge, Mass: Harvard University Press.

19.

Grosjean

(2008). Studying bilinguals. Oxford, UK: Oxford University Press.

20.

Holm

(1979). A simple sequentially rejective multiple test procedure. Scandinavian Journal of Statistics, 6, 65–70.

21.

Horwitz

(2010). Foreign and second language anxiety. Language Teaching, 43, 154–167.

22.

Kim

K. H.

Bang

S. W.

Kim

S. R.

(2004). Emotion recognition system using short-term monitoring of physiological signals. Medical and Biological Engineering and Computing, 42(3), 419–427.

23.

Knight

M. L.

Borden

R. J.

(1979). Autonomic and affective reactions of high and low socially-anxious individuals awaiting public performance. Psychophysiology, 16, 209–213.

24.

Kreibig

S. D.

(2010). Autonomic nervous system activity in emotion: a review. Biological Psychology, 84, 394–421.

25.

MacIntyre

P. D.

(1999). Language anxiety: a review of the research for language teachers. In Young

D. J.

(Ed.), Affect in foreign language and second language learning: a practical guide to creating a low-anxiety classroom atmosphere (pp. 24–45). Boston: McGraw-Hill.

26.

MacIntyre

P. D.

(2017). An overview of language anxiety research and trends in its development. In Gkonou

Daubney

Dewaele

J-M.

(Eds.), New insights into language anxiety: Theory, research and educational implications (pp. 11–30). Bristol: Multilingual Matters.

27.

MacIntyre

P. D.

Noels

K. A.

Clément

(1997). Biases in self-ratings of second language proficiency: the role of language anxiety. Language Learning, 47, 265–287.

28.

MacIntyre

V. A.

MacIntyre

P. D.

Carre

(2010). Heart rate variability as a predictor of speaking anxiety. Communication Research Reports, 27, 286–297.

29.

Moore

D. L.

Baron

R. S.

(1983). Social facilitation: A psychophysiological analysis. In Cacioppo

J. T.

Petty

R. E.

(Eds.), Social psychophysiology: A sourcebook (pp. 434–466). New York: Guilford.

30.

Rothman

Treffers-Daller

(2014). A prolegomenon to the construct of the native speaker: heritage speaker bilinguals are natives too! Applied Linguistics, 35, 93–98.

31.

Santos

Cenoz

Gorter

(2017). Communicative anxiety in English as a third language. International Journal of Bilingual Education and Bilingualism, 20(7), 823–236.

32.

Sevinç

(2016). Language maintenance and shift under pressure: Three generations of the Turkish immigrant community in the Netherlands. International Journal of the Sociology of Language, 242, 81–117.

33.

Sevinç

(2017). Language anxiety in the immigrant context. An interdisciplinary perspective (doctoral dissertation). University of Oslo, Oslo, Norway.

34.

Sevinç

Dewaele

J.-M.

(2018). Heritage language anxiety and majority language anxiety among Turkish immigrants in the Netherlands. International Journal of Bilingualism, 22(2), 159–179.

35.

Spielberger

C. D.

(1972). Anxiety as an emotional state. In Speilberger

C. D.

(Ed.), Anxiety: current trends in theory and research (pp. 23–49). New York: Academic Press.

36.

Steinberg

F. S.

Horwitz

E. K.

(1986). The effect of induced anxiety on the denotative and interpretative content of second language speech. TESOL Quarterly, 20(1), 131–136.

37.

Valdés

(2005). Bilingualism, heritage language learners, and SLA research: opportunities lost or seized? Modern Language Journal, 89(3), 410–426.

38.

Witt

P. L.

Brown

K. C.

Roberts

J. B.

Weisel

Sawyer

C. R.

Behnke

R. R.

(2006). Somatic anxiety patterns before, during, and after giving a public speech. Southern Communication Journal, 71, 87–100.