Stress and accent in community interpreting

3.1 Materials

The materials comprised two job interviews (interview A and interview B) between an interviewer speaking English and an interviewee speaking Spanish. Job interviews were selected as an encounter that would be suited for a community interpreter since interviews contain prototypical community interpreting interactions that follow a question-and-answer discourse and relatively short exchanges between interlocutors. Moreover, these encounters are likely to be between two parties who have not previously met or interacted, such that the interpreter is unlikely to draw on previous knowledge of the speakers when completing the tasks.

The interviews were of equal length (interview A – 899 words, including 433 words in the interviewer’s part, 6 min and 9 s; interview B – 899 words, including 431 words in the interviewer’s part, 6 min and 7 s). The materials contained no specialized terminology, affect-laden language, or other problem triggers such as complex numbers. The interviews also followed the same structure: opening, a series of interview questions, a closing part including information on how feedback will be provided to the interviewee. For comparability, both interviews were for the position of Senior Manager in the banking industry.

Two versions of each interview were prepared: one with a male English speaker using Midwestern American English (sometimes referred to as General American English or Standard American English), and the other with a male English speaker using Southern American English, a regional accent variety of English with linguistic features of Southern Midland English.^[1] Midwestern American English is often considered the mainstream variety of English spoken by most Americans and is not conceptualized as being a single language accent (Kövecses 2000), while Southern Midland English is more characteristic of speakers inland in the Missouri, Arkansas, and Tennessee regions. The latter share similar vowel patterns as Southern accents in the United States along with sonorant consonants (Labov et al. 2006). Both speakers are native speakers of English, were educated in the United States, and hold doctoral degrees. Each has lived for extended periods of time in the respective regions where each language variety is spoken. The same female speaker recorded the Spanish-language interviewee’s parts for both interviews. Speakers were told to speak at their regular rate of speed and were not instructed to change their manner of speech for the recording.

3.2 Participants

A total of 18 interpreters took part in the study via purposive sampling. They were recruited through professional interpreting listservs and invitations sent by email in order to specifically identify interpreters working in community settings. Interpreting students were also recruited via email by instructors currently or previously teaching community interpreting. To participate in the study, they needed to be at least 18 years old, be either an interpreting student or a professional community interpreter, and have some experience interpreting in the English-Spanish language pair. Experience was not controlled as an inclusion criterion in order to provide a sufficiently broad range of participants in the study – so long as the participants had interpreting experience in the Spanish-English language pair and met the other inclusion criteria, they were not disqualified from the study. In the final sample, 9 were US-based interpreting trainees and the other 9 were US-based professional community interpreters. The sample included 3 men and 15 women, ranging in age from 20 to 69.^[2] At the time of participation, interpreting trainees had 1–2 years of translation/interpreting education. All professional interpreters were certified (either court or medical interpreters). All participants had English and Spanish as their primary interpreting language combination. 14 reported formal education in Spanish, while 16 had experience living in a Spanish-speaking country. Given the challenges associated with identifying ‘native language’ or ‘L1’ in relation to language proficiency in interpreting and using these as potential demographic indicators (for a discussion, see Tiselius 2025), participants were instead asked to self-assess their language proficiency in English and Spanish on a 1–5 scale (1 = ‘none’; 5 = ‘native-like or near-native’) in four language-related abilities: speaking, listening, reading, and writing. Participants rated their language proficiency in English as follows: speaking: M = 4.94, SD = 0.24; listening: M = 4.88, SD = 0.33; reading: M = 4.94, SD = 0.24; writing: M = 4.94, SD = 0.24, while in Spanish the ratings were as follows: speaking: M = 4.61, SD = 0.5; listening: M = 4.82, SD = 0.39; reading: M = 4.65, SD = 0.49; writing: M = 4.53, SD = 0.62. This more granular information provides greater insight since phonological awareness and perception across accents cannot be fully accounted for with native-language indicators alone (see Colina 2025). Since one participant did not provide answers for listening, reading, and writing, the results in these categories are based on 17 responses, while the results for speaking were calculated for all 18 participants.

3.3 Measurement

The construct of self-efficacy was considered in this study so as to look for potential relationships with interpreters’ self-reported stress. As self-efficacy has been shown to modulate a physiological stress response (e.g., O’Leary 1992), with this study we aimed to test whether a similar relationship can be observed for self-reported stress. To this end, the latent construct of interpreter self-efficacy was measured utilizing the Interpreting Self-Efficacy (ISE) scale (Lee 2014). ISE is a 21-item scale with three sub-dimensions of Self-Confidence (SC), Self-Regulatory Efficacy (SR), and Preference for Task Difficulty (TD; Lee 2014), which aligns with Kim and Park’s (2001) model of academic self-efficacy. Responses to the scale’s items are provided on a 6-point Likert-type scale in line with its original development and implementation. We calculated values for both general self-efficacy and separately for the three sub-dimensions using the following scale: “definitely true of me” = 1, “true of me” = 2, “somewhat true of me” = 3, “somewhat untrue of me” = 4, “untrue of me” = 5, and “very untrue of me” = 6; because these linguistic markers are the reverse of Lee’s (2014) original response scale, the scores were reversed to allow interpretability and comparability to previous research. Additionally, reverse scoring was applied for 5 items, based on Lee’s (2014) instructions. Since the original scale was designed with the aim of measuring student interpreters’ self-efficacy, the language for professional interpreters taking part in this study was slightly edited to consider the participants’ professional status. For example, “other students” was replaced with “other professional interpreters” and “In an interpretation classroom” was changed to “When interpreting.”

Stress experienced during the task was measured using the Short Stress State Questionnaire (SSSQ; Helton 2004). The SSSQ is a 24-item tool used to assess subjective state stress as a result of a performed task; it comprises three main factors: task engagement, distress, and worry, and answers are provided on a 5-point Likert scale. The tool features two versions: one to measure perceived stress before the task (State Pre-Questionnaire) and after it is completed (State Post-Questionnaire).

3.4 Procedure

The study was approved by the Institutional Review Board of UNC Charlotte [approval no. 21-0332]. It involved interpreting two audio recordings of job interviews consecutively between an English speaker and a Spanish speaker. The study was conducted via Zoom, and each experimental session took approximately 60 min. All survey data were collected using Qualtrics. Informed consent was obtained electronically via DocuSign before participation in the experiment. Participants could withdraw from the study at any time, and no incentive was provided for participation in the study.

Prior to the start of the experimental tasks, participants completed three online surveys: a demographic questionnaire, Interpreting Self-Efficacy (ISE) scale, and the Short Stress State Pre-Questionnaire (SSSQ). Each participant was assigned a participant number to allow the data from each of the experimental tasks to be matched throughout the experiment. No identifiable data were collected so participant identities could not be determined after the experimental tasks were completed.

In the main part of the experiment, each participant interpreted two job interviews differing in the accent of the English speaker. The order of presentation (interview A vs. interview B) and the version provided (general accent vs. regional accentual variety) were counterbalanced across the participants. Interviews were interpreted consecutively in short fragments – 33 interpreting chunks in both interviews. The average number of words per chunk equaled 27.2 for both interviews. The recording was managed by the experimenter, who stopped the recording after each fragment so that an interpretation could be provided. Participants were instructed that they could take notes if they wished.

After each interpreting task the Short Stress State Post-Questionnaire (SSSQ; Helton 2004) was completed by the participant to test the impact of the experimental condition on perceived stress. Short semi-structured interviews were also conducted to collect qualitative data. After each task, participants were asked about their impressions of the task, potential challenges, and perceived stress. After both interpretations were completed, participants were also asked to compare the tasks regarding difficulty, the speakers’ accent, the level of perceived stress, and physical manifestations of the stress experienced (see Appendix for the list of post-task and post-experiment questions). At the end of the experiment, participants were debriefed about the study’s aims and hypotheses.

3.5 Data analysis

The scores for the Interpreting Self-Efficacy (ISE) scale (Lee 2014) were analyzed for three factors, i.e., Self-Confidence, Self-Regulatory Efficacy, and Preference for Task Difficulty (Lee 2014). Scores were calculated for each subscale (Self-Confidence – 9 items; Self-Regulatory Efficacy – 4 items; Preference for Task Difficulty – 8 items), as well as for all 21 items as a measure of overall self-efficacy.

Participants’ stress level was measured with the Short Stress State Questionnaire (SSSQ; Helton 2004). The pre-task levels were treated as a point of reference, and for both interpreting tasks, the measured SSSQ score was divided by the baseline level. Thus, the calculated values measured how the level of perceived stress changed as a function of the interpreting task completed. Scores were calculated separately for the three SSSQ subscales: task engagement, distress, and worry. As a robustness check, all analyses were repeated using the difference (rather than the ratio) between baseline and experimental conditions; results were qualitatively similar.

Correlation analysis was employed to test the relationship between self-efficacy and self-reported stress. The effect of accent on stress was tested with repeated measures ANOVA and paired t-tests. Data analysis was conducted using R software.

4.1 Stress and self-efficacy correlation

When analyzing responses to the ISE, participants generally reported moderate to strong self-efficacy with the following summary statistics: self-confidence (SC): M = 4.94, SD = 0.57; self-regulatory efficacy (SR): M = 5.07, SD = 0.59; preference for task difficulty (TD): M = 4.12, SD = 0.84; total ISE: M = 4.65, SD = 0.60. The scale’s reliability was strong for SC (Cronbach’s α = 0.89, 95 % CI [0.82, 0.97]) and TD (Cronbach’s α = 0.86, 95 % CI [0.77, 0.96]). Reliability for SR was a bit weaker with a wide confidence interval (Cronbach’s α = 0.63, 95 % CI [0.32, 0.94]), in part likely due to the smaller number of items for the subscale.

Pearson’s correlations were computed between the ISE and SSSQ measures in both the accented and unaccented condition, and results appear in Table 1. The magnitude of all correlations is quite small, with none of them statistically significant at the 5 % level. Therefore, the relationship observed between stress and self-efficacy was small for the participants in this study (Mellinger and Hanson 2017).

4.2 Self-reported stress

The potential effect of accent on self-reported stress was tested with three repeated measures ANOVA models, one for each subscale of the SSSQ. This omnibus test helps control for Type I error. The primary factor of interest was accent, but the model also included professional status (professional or student), the experimental order (accented or unaccented version first), and the interaction between accent and experimental order.

Accent was not statistically significant at the 5 % level in any of the three models: worry, F[1] = 0.003, p = 0.957; engagement, F[1] = 0.201, p = 0.657; distress, F[1] = 3.58, p = 0.068. Neither professional status nor the experimental order was statistically significant in any of the three models. However, there was a statistically significant interaction between order effect and accent for the distress dimension of the SSSQ (F[1] = 4.80, p = 0.037).

While the omnibus tests were not statistically significant, results of paired t-tests for accent and order effect allow for clearer presentation of the descriptive statistics and effect sizes. For the distress subscale, there was no statistically significant difference between the accented (M = 1.34, SD = 0.46) and non-accented (M = 1.27, SD = 0.34) conditions with a small effect size (t[17] = 0.56, p = 0.582, Cohen’s d = 0.173). Results were similarly not statistically significant for the engagement subscale when comparing the accented (M = 0.94, SD = 0.16) and non-accented (M = 0.95, SD = 0.15) conditions with a small effect size (t[17] = 0.21, p = 0.836, Cohen’s d = 0.06). Finally, there was no statistically significant difference observed between the accented (M = 1.06, SD = 0.47) and non-accented (M = 1.07, SD = 0.48) conditions for the worry subscale of the SSSQ with a small effect size (t[17] = 0.26, p = 0.801, Cohen’s d = 0.02). Overall, the change in self-reported stress was negligible when comparing the accented and non-accented versions.

In order to determine whether there was a task effect present, we also incorporated task order as part of the overall analytical model. Experimental order did not have a strong influence on self-reported stress either. The results of t-tests include descriptive statistics and effect sizes. For the distress subscale of the SSSQ there was no statistically significant difference between the accented-first (M = 1.42, SD = 0.50) and non-accented-first (M = 1.22, SD = 0.29) conditions with a medium effect size (t[22.8] = 1.41, p = 0.171, Cohen’s d = 0.49). There was also no statistically significant difference for the engagement subscale between the accented-first (M = 0.90, SD = 0.16) and non-accented-first (M = 0.98, SD = 0.14) conditions with a medium effect size (t[29.3] = 0.21, p = 0.129, Cohen’s d = 0.53). Finally, no statistically significant difference was observed between the accented-first (M = 1.14, SD = 0.62) and non-accented-first (M = 1.01, SD = 0.30) conditions for the worry subscale with a small effect size (t[20.5] = 0.77, p = 0.452, Cohen’s d = 0.27).

The one-dimensional analyses confirm the omnibus ANOVA results that neither accent nor experimental order are independently associated with statistically significant differences in self-reported stress levels, though the effect sizes related to task order are stronger. However, the one-way ANOVA did reveal an interaction effect of these two dimensions that suggests experimental order plays at least some role. The interaction effect was statistically significant for the distress scale. Descriptive statistics, partitioned by both accent and experimental order provide further insight, as seen in Table 2. Whether participants began with the accented version or the non-accented version, in both situations the second task was reported as less stressful, which results in the observed interaction effect. The pattern is similar for the engagement and worry subdimensions of the SSSQ, though the results are not statistically significant. These results reinforce that accent is not a primary driver of participant stress, but there may be some evidence of a warm-up effect such that participants become more comfortable with the second task. The difference could also arise from increasing comfort with the experimental setup or a combination of the two effects.

4.3 Semi-structured interviews

Semi-structured interviews were conducted after each task to collect qualitative data about participants’ impressions about the tasks, potential difficulties involved, and perceived stress. At the end of the experiment, participants were also asked questions that aimed to compare the tasks in terms of perceived difficulty, the speakers’ accent, the level of perceived stress, including its physical manifestations. Transcripts of interview questions were prepared and then annotated for recurring topics. These topics referred to their perception of the tasks, the challenges involved and the stress experienced while performing the tasks.

One of the recurrent topics in the participants’ answers was task immersion, that is, how involved participants were in an interpreting task. Given the nature of the interpreted events (i.e., job interviews), participants reported the responsibility they felt for the interpreted content. Even though they were aware that they were involved in a research project, some of them said that the task was stressful as making errors could result in the candidate not being able to secure the position she was applying for.

Actually, yes, I found it very stressful in the moment because I thought of it as I’m interpreting for these people in real life. (Participant 24)

I would say difficult because the stakes are so high (…). This is hiring a senior manager. And like, you know, I don’t want to blow it for her, and I don’t want him to get a wrong impression there. (Participant 2)

So I don’t think it’s possible to interpret without any stress at all, because we’re always in important settings where things matter to people. (Participant 4)

Participants were also asked questions related to stress while interpreting. In terms of stress factors, some participants referred to some cognitive challenges that triggered a stressful reaction, such as the need to memorize the content of the chunks (Participants 5, 24, and 35), information overload (Participant 25), and fast delivery rate (Participant 5). When asked about bodily manifestations of stress, participants provided the following examples:

I found myself playing with a pen or trying to take notes and just moving around a lot with my hands. (Participant 24)

I noticed, especially at the beginning, my heart rate was really fast. I really noticed that. (Participant 2)

There is more tension in the body when stress increases. (Participant 3)

When I began I crossed my legs underneath me, which I sometimes do when I’m feeling nervous. (Participant 20)

I just found myself like shaking my head like feeling a little frustrated with myself. (Participant 21)

Psychological literature recognizes both negative (distress) and positive (eustress) aspects of stress (e.g., Selye 1974). This distinction was also visible in the answers to interview questions provided by participants. While some participants recognized that the experience of stress may be perceived as unpleasant and trigger anxiety, they also appreciated the positive aspects of stress, as exemplified by the following two passages:

I feel invigorated when I have to try to kind of work around things. And I also felt like I needed to make sure I wasn’t falling into any of the traps of false cognates, because I felt like there were several. (Participant 6)

It’s just the typical adrenaline rush. So being put on the spot. (Participant 3)

Accented speech was hypothesized to serve as a potential stress factor in this study. However, a quantitative analysis of the Short Stress State Questionnaire (SSSQ; Helton 2004) scores showed that an accented condition was not more stressful than interpreting a non-accented interview at a statistically significant level. This result may result from US interpreters’ exposure to various accents of both English and Spanish, a factor that was also mentioned in some interviews. Some participants reported that since they have experience hearing and interpreting many accentual varieties of English or Spanish, the content that they interpreted was not a problematic factor that could affect their performance.

It was American English and I’ve been in the US for 10 years and that’s the American, that’s the English that I’m used to. So it wasn’t challenging at all in that sense. (Participant 23)

I live in that, in like the southern US. I’m kind of already accustomed to that. Some of my co-workers talk like that and sometimes I catch myself speaking like that. So that was easier. (Participant 17)

I lived in Spain for, for six years so I’m very used to hearing Spanish accents. (Participant 2)

Analysis of interviews pointed to two pertinent methodological considerations. First, when studying stress in the context of interpreting, there may be a confluence of stress resulting from experimental manipulation (independent variable) and stress triggered by a test situation. In other words, what might have increased the level of stress experienced by participants in this study is not only accented speech, but also the fact that participants might have been anxious about their performance being thoroughly examined and quantified in a research study setting. This potential observer effect is noted by Participant 1:

Yes, I found it stressful. Just, you know, like the testing situation. (Participant 1)

Second, in their answers many participants discussed the effects of exposure to an interpreting task. For the sake of experimental control, interviews were carefully matched for the type of interpreted event, topic, length, and the level of complexity. Having interpreted the first interview, participants could have anticipated the content of the interpreted event and potential challenges involved. These conclusions inspired us to test the potential impact of task order on stress experienced by participants, which is supported by the analysis of the SSSQ scores. The potential role of task order can be discussed by looking at the following excerpts from interviews:

I feel like I was more confident in my interpretation and kind of could anticipate more what was coming. (Participant 8)

I feel like the first exercise was more stressful and I don’t know if it’s because it was totally unfamiliar. And the second one was similar, and so I had some basis of knowledge for it. It wasn’t the first time that I had heard it. So I had time to think a little bit about some of the terms. (Participant 9)

I believe the first one felt more difficult because it was my first encounter, and I got a bit more comfortable after the first one. And I was a bit less nervous so yeah the first one did feel a bit more difficult. (Participant 22)

However, Participant 20 suggested that the second task might have been more stressful as a result of accumulated stress from both tasks:

It felt like a compounding effect from what little stress I had accrued from the first task. (Participant 20)

In summary, semi-structured interviews were conducted to triangulate quantitative data from psychometric tools with a qualitative approach. Answers provided by the participants offered a more nuanced insight into stress experienced by community interpreters along with its physiological manifestations during the experimental task. Answers to interview questions also offered a plausible explanation of the obtained results. Lack of statistically significant differences between the accented and non-accented condition may result from US interpreters’ extensive exposure to accents in English and Spanish. Answers provided by the participants also appear to support the SSSQ results suggesting a possible practice effect (stress experienced in response to first vs. second task).