Disentangling cognitive and emotional load in interpreting: Multimodal evidence from physiological, acoustic, and self-report measures

Alper Kumcu and Duygu Duman
Hacettepe University | Bilkent University

This study employs a multimodal approach to investigate how emotional and cognitive load differentially impact interpreter stress. Twenty student interpreters consecutively interpreted four Turkish-to-English mental healthcare narratives of varying emotionality and difficulty, constructed using a 2×2 factorial design. Stress responses were measured using galvanic skin response (GSR), voice parameters (pitch and voice intensity), and self-report scales (PANAS, STAI, and NASA-TLX). Stimuli were normed for emotional valence and linguistic complexity using readability indices, expert ratings, and natural language processing (NLP) tools. Results showed that primarily cognitive load drove GSR arousal, particularly during later interpreting stages, whereas emotional load was reflected in pitch and self-reported negative affect. A pitch convergence between speaker and interpreter also emerged. None of the stress indicators predicted interpreting accuracy, suggesting a possible role for individual coping strategies.

Publication history

Date received: 17 July 2025

Date accepted: 12 October 2026

Published online: 5 November 2026

Table of contents

1.Introduction

Stress is widely regarded as one of the most critical predictors of an interpreter’s well-being and performance. Data from surveys and field interviews (Daly and Chovaz 2020Daly, Brad, and Cathy J. Chovaz 2020 “Secondary Traumatic Stress: Effects on the Professional Quality of Life of Sign Language Interpreters.” American Annals of the Deaf 165 (3): 353–367. Google Scholar logo with link to Google ScholarDaly, Brad, and Cathy J. Chovaz 2020 “Secondary Traumatic Stress: Effects on the Professional Quality of Life of Sign Language Interpreters.” American Annals of the Deaf 165 (3): 353–367. Google Scholar logo with link to Google Scholar; Simms et al. 2021Simms, James V., Alexandra R. Thelan, Sarah E. Domoff, and Elizabeth A. Meadows 2021 “An Examination of Vicarious Trauma among Refugee Mental Health Interpreters.” Occupational Health Science 5 (4): 581–601. Google Scholar logo with link to Google ScholarSimms, James V., Alexandra R. Thelan, Sarah E. Domoff, and Elizabeth A. Meadows 2021 “An Examination of Vicarious Trauma among Refugee Mental Health Interpreters.” Occupational Health Science 5 (4): 581–601. Google Scholar logo with link to Google Scholar; Walczyński 2021Walczyński, Marcin 2021 “Polish–English Certified Interpreters in Psycho-Affectively Challenging Contexts.” Current Trends in Translation Teaching and Learning E 8: 394–448. Google Scholar logo with link to Google ScholarWalczyński, Marcin 2021 “Polish–English Certified Interpreters in Psycho-Affectively Challenging Contexts.” Current Trends in Translation Teaching and Learning E 8: 394–448. Google Scholar logo with link to Google Scholar) have repeatedly shown that community interpreting, especially healthcare interpreting, involves the most challenging communication scenarios, testing interpreters’ resilience (Park et al. 2017Park, Elyse R., Jan E. Mutchler, Giselle Perez, et al. 2017 “Coping and Resiliency Enhancement Program (CARE): A Pilot Study for Interpreters in Cancer Care.” Psycho-Oncology 26 (8): 1181–1190. Google Scholar logo with link to Google ScholarPark, Elyse R., Jan E. Mutchler, Giselle Perez, et al. 2017 “Coping and Resiliency Enhancement Program (CARE): A Pilot Study for Interpreters in Cancer Care.” Psycho-Oncology 26 (8): 1181–1190. Google Scholar logo with link to Google Scholar). Consequently, community interpreters are particularly susceptible to vicarious traumatisation, which can lead to long-term psychological effects such as post-traumatic stress disorder, depressive symptoms, anxiety, and burnout (Kindermann et al. 2017Kindermann, David, Carolin Schmid, Cassandra Derreza-Greeven, et al. 2017 “Prevalence of and Risk Factors for Secondary Traumatization in Interpreters for Refugees: A Cross-Sectional Study.” Psychopathology 50 (4): 262–272. Google Scholar logo with link to Google ScholarKindermann, David, Carolin Schmid, Cassandra Derreza-Greeven, et al. 2017 “Prevalence of and Risk Factors for Secondary Traumatization in Interpreters for Refugees: A Cross-Sectional Study.” Psychopathology 50 (4): 262–272. Google Scholar logo with link to Google Scholar); negative feelings such as being overwhelmed, experiencing guilt, and feeling lonely and uncomfortable (Hancox et al. 2023Hancox, Jennifer A., Clare F. McKiernan, Alice L. Martin, Jon Tomas, and John I. MacArtney 2023 “The Emotional Effects on Professional Interpreters of Interpreting Palliative Care Conversations for Adult Patients: A Rapid Review.” Palliative Medicine 37 (7): 931–946. Google Scholar logo with link to Google ScholarHancox, Jennifer A., Clare F. McKiernan, Alice L. Martin, Jon Tomas, and John I. MacArtney 2023 “The Emotional Effects on Professional Interpreters of Interpreting Palliative Care Conversations for Adult Patients: A Rapid Review.” Palliative Medicine 37 (7): 931–946. Google Scholar logo with link to Google Scholar); and adverse occupational outcomes such as absenteeism and high staff turnover (Lai and Heydon 2015Lai, Miranda, and Georgina Heydon 2015 “Vicarious Trauma among Interpreters.” International Journal of Interpreter Education 7 (1): 3–23.Google Scholar logo with link to Google ScholarLai, Miranda, and Georgina Heydon 2015 “Vicarious Trauma among Interpreters.” International Journal of Interpreter Education 7 (1): 3–23.Google Scholar logo with link to Google Scholar; Delizée et al. 2024Delizée, Anne, Nathalie Bennoun, Felicia Dutray, Lorine Pierard, and Orest Weber 2024 “Preventing Vicarious Traumatisation in Mental Health Settings: A Training Course for Interpreters.” FITISPos International Journal 11 (1): 28–43. Google Scholar logo with link to Google ScholarDelizée, Anne, Nathalie Bennoun, Felicia Dutray, Lorine Pierard, and Orest Weber 2024 “Preventing Vicarious Traumatisation in Mental Health Settings: A Training Course for Interpreters.” FITISPos International Journal 11 (1): 28–43. Google Scholar logo with link to Google Scholar). Against this backdrop, the present study sought to explore stress, its origins, and its manifestations in consecutive interpreting (CI). Participants consecutively interpreted four narratives about mental healthcare from their L1 to L2, varying in cognitive and emotional intensity. Their skin responses and speech outputs were recorded while they performed the task. They self-reported their affect after the tasks, and we analysed their performance in interpreting.

‘Emotion’ is often used as an umbrella term encompassing related constructs such as affect, stress, and anxiety. Since each of these constructs corresponds to distinct measures in the present study, it is essential to define them clearly from the outset. ‘Affect’ refers to a person’s core affective state, a continuously varying neurophysiological condition characterised by valence (ranging from pleasant to unpleasant) and arousal (from high to low activation). It constitutes the fundamental experiential dimension underlying all emotional and stress-related reactions. ‘Stress’ and ‘anxiety’ are built upon these affective responses when they are interpreted or conceptualised in specific ways. In this respect, stress arises from a real or perceived stressor or imminent threat and prepares the organism for action. It invariably entails physiological activation, including increased heart rate, respiratory rate, cortisol levels, and/or muscle tension. It may be experienced as either adaptive (proactive) or maladaptive (avoidant), depending on whether the arousal is interpreted as a challenge or a threat. Anxiety shares the same physiological base as stress but is typically anticipatory, emerging from the perception or expectation of threat even in the absence of an external stressor. Consequently, anxiety reflects how individuals interpret bodily arousal, such as worry, fear, or tension, based on prior experiences (Rojo López and Foulquié Rubio 2024Rojo López, Ana María, and Ana Isabel Foulquié Rubio 2024 “Interpreting, Affect, and Emotion.” In The Routledge Handbook of Interpreting and Cognition, edited by Christopher D. Mellinger, 307–323. London: Routledge. Google Scholar logo with link to Google ScholarRojo López, Ana María, and Ana Isabel Foulquié Rubio 2024 “Interpreting, Affect, and Emotion.” In The Routledge Handbook of Interpreting and Cognition, edited by Christopher D. Mellinger, 307–323. London: Routledge. Google Scholar logo with link to Google Scholar). When coping resources are insufficient to meet the cognitive and psychological demands of the environment, stress and anxiety can manifest in physical, psychological, or social dysfunctions (Richardson and Rothstein 2008Richardson, Katherine M., and Hannah R. Rothstein 2008 “Effects of Occupational Stress Management Intervention Programs: A Meta-Analysis.” Journal of Occupational Health Psychology 13 (1): 69–93. Google Scholar logo with link to Google ScholarRichardson, Katherine M., and Hannah R. Rothstein 2008 “Effects of Occupational Stress Management Intervention Programs: A Meta-Analysis.” Journal of Occupational Health Psychology 13 (1): 69–93. Google Scholar logo with link to Google Scholar). Both stress and anxiety are thus particularly salient in interpreting settings, where interpreters must manage intense cognitive load and emotional regulation demands, often with consequences for well-being and performance (Riccardi, Marinuzzi, and Zecchin 1998Riccardi, Alessandra, Guido Marinuzzi, and Stefano Zecchin 1998 “Interpretation and Stress.” The Interpreters’ Newsletter 8: 93–106.Google Scholar logo with link to Google ScholarRiccardi, Alessandra, Guido Marinuzzi, and Stefano Zecchin 1998 “Interpretation and Stress.” The Interpreters’ Newsletter 8: 93–106.Google Scholar logo with link to Google Scholar).

Investigations into stress within the broader field of psycho-affective research in interpreting are gaining ground, resulting in a ‘psycho-affective turn’ in Interpreting Studies (Li, Wang, and Rasmussen 2022Li, Saihong, Yifang Wang, and Yubo Zhou Rasmussen 2022 “Studying Interpreters’ Stress in Crisis Communication: Evidence from Multimodal Technology of Eye-Tracking, Heart Rate and Galvanic Skin Response.” The Translator 28 (4): 468–488. Google Scholar logo with link to Google ScholarLi, Saihong, Yifang Wang, and Yubo Zhou Rasmussen 2022 “Studying Interpreters’ Stress in Crisis Communication: Evidence from Multimodal Technology of Eye-Tracking, Heart Rate and Galvanic Skin Response.” The Translator 28 (4): 468–488. Google Scholar logo with link to Google Scholar). Furthermore, researchers are making use of a growing range of methodological tools to detect and measure the psychophysiological effects of stress in interpreting situations (Korpal 2016Korpal, Paweł 2016 “Interpreting as a Stressful Activity: Physiological Measures of Stress in Simultaneous Interpreting.” Poznan Studies in Contemporary Linguistics 52 (2): 297–316. Google Scholar logo with link to Google ScholarKorpal, Paweł 2016 “Interpreting as a Stressful Activity: Physiological Measures of Stress in Simultaneous Interpreting.” Poznan Studies in Contemporary Linguistics 52 (2): 297–316. Google Scholar logo with link to Google Scholar; Gieshoff, Lehr, and Hunziker Heeb 2021Gieshoff, Anne Catherine, Caroline Lehr, and Andrea Hunziker Heeb 2021 “Stress, Cognitive, Emotional and Ergonomic Demands in Interpreting and Translation: A Review of Physiological Studies.” Cognitive Linguistic Studies 8 (2): 404–439. Google Scholar logo with link to Google ScholarGieshoff, Anne Catherine, Caroline Lehr, and Andrea Hunziker Heeb 2021 “Stress, Cognitive, Emotional and Ergonomic Demands in Interpreting and Translation: A Review of Physiological Studies.” Cognitive Linguistic Studies 8 (2): 404–439. Google Scholar logo with link to Google Scholar; Muñoz Martín and Rojo López 2025Muñoz Martín, Ricardo, and Ana María Rojo López 2025 “Introduction: Research Methods in Cognitive Translation and Interpreting Studies.” In Research Methods in Cognitive Translation and Interpreting Studies, edited by Ana María Rojo López and Ricardo Muñoz Martín, 1–20. Amsterdam: John Benjamins. Google Scholar logo with link to Google ScholarMuñoz Martín, Ricardo, and Ana María Rojo López 2025 “Introduction: Research Methods in Cognitive Translation and Interpreting Studies.” In Research Methods in Cognitive Translation and Interpreting Studies, edited by Ana María Rojo López and Ricardo Muñoz Martín, 1–20. Amsterdam: John Benjamins. Google Scholar logo with link to Google Scholar). These tools include self-reports of affective states (Gumul 2021Gumul, Ewa 2021 “Reporting Stress in Simultaneous Interpreting: The Analysis of Trainee Interpreters’ Retrospective Reports and Outputs.” Onomázein NE8: 16–42.Google Scholar logo with link to Google ScholarGumul, Ewa 2021 “Reporting Stress in Simultaneous Interpreting: The Analysis of Trainee Interpreters’ Retrospective Reports and Outputs.” Onomázein NE8: 16–42.Google Scholar logo with link to Google Scholar); heart/pulse rate (Klonowicz 1994 1994 “Putting One’s Hart into Simultaneous Interpretation.” In Bridging the Gap: Empirical Research in Simultaneous Interpretation, edited by Sylvie Lambert and Barbara Moser-Mercer, 213–224. Amsterdam: John Benjamins. Google Scholar logo with link to Google Scholar 1994 “Putting One’s Hart into Simultaneous Interpretation.” In Bridging the Gap: Empirical Research in Simultaneous Interpretation, edited by Sylvie Lambert and Barbara Moser-Mercer, 213–224. Amsterdam: John Benjamins. Google Scholar logo with link to Google Scholar; Olalla-Soler, Spinolo, and Muñoz Martín 2023Olalla-Soler, Christian, Nicoletta Spinolo, and Ricardo Muñoz Martín 2023 “Under Pressure? A Study of Heart Rate and Heart-Rate Variability Using SmarTerp.” HERMES: Journal of Language and Communication in Business 63: 119–142.Google Scholar logo with link to Google ScholarOlalla-Soler, Christian, Nicoletta Spinolo, and Ricardo Muñoz Martín 2023 “Under Pressure? A Study of Heart Rate and Heart-Rate Variability Using SmarTerp.” HERMES: Journal of Language and Communication in Business 63: 119–142.Google Scholar logo with link to Google Scholar); blood pressure (Roziner and Shlesinger 2010Roziner, Ilan, and Miriam Shlesinger 2010 “Much Ado about Something Remote: Stress and Performance in Remote Interpreting.” Interpreting 12 (2): 214–247. Google Scholar logo with link to Google ScholarRoziner, Ilan, and Miriam Shlesinger 2010 “Much Ado about Something Remote: Stress and Performance in Remote Interpreting.” Interpreting 12 (2): 214–247. Google Scholar logo with link to Google Scholar); cortisol levels (Moser-Mercer 2005Moser-Mercer, Barbara 2005 “Remote Interpreting: The Crucial Role of Presence.” Bulletin VALS-ASLA 81: 73–97.Google Scholar logo with link to Google ScholarMoser-Mercer, Barbara 2005 “Remote Interpreting: The Crucial Role of Presence.” Bulletin VALS-ASLA 81: 73–97.Google Scholar logo with link to Google Scholar); electromyography (EMG) (Delisle et al. 2005Delisle, Alain, Christian Larivière, Daniel Imbeau, and Marie-José Durand 2005 “Physical Exposure of Sign Language Interpreters: Baseline Measures and Reliability Analysis.” European Journal of Applied Physiology 94 (4): 448–460. Google Scholar logo with link to Google ScholarDelisle, Alain, Christian Larivière, Daniel Imbeau, and Marie-José Durand 2005 “Physical Exposure of Sign Language Interpreters: Baseline Measures and Reliability Analysis.” European Journal of Applied Physiology 94 (4): 448–460. Google Scholar logo with link to Google Scholar); galvanic skin response (GSR) (Kurz 2002Kurz, Ingrid 2002 “Physiological Stress Responses during Media and Conference Interpreting.” In Interpreting in the 21st Century: Challenges and Opportunities, edited by Giuliana Garzone and Maurizio Viezzi, 195–202. Amsterdam: John Benjamins. Google Scholar logo with link to Google ScholarKurz, Ingrid 2002 “Physiological Stress Responses during Media and Conference Interpreting.” In Interpreting in the 21st Century: Challenges and Opportunities, edited by Giuliana Garzone and Maurizio Viezzi, 195–202. Amsterdam: John Benjamins. Google Scholar logo with link to Google Scholar, 2003 2003 “Physiological Stress during Simultaneous Interpreting: A Comparison of Experts and Novices.” The Interpreters’ Newsletter 12: 51–67.Google Scholar logo with link to Google Scholar 2003 “Physiological Stress during Simultaneous Interpreting: A Comparison of Experts and Novices.” The Interpreters’ Newsletter 12: 51–67.Google Scholar logo with link to Google Scholar; Korpal and Jasielska 2019Korpal, Paweł, and Aleksandra Jasielska 2019 “Investigating Interpreters’ Empathy: Are Emotions in Simultaneous Interpreting Contagious?” Target 31 (1): 2–24. Google Scholar logo with link to Google ScholarKorpal, Paweł, and Aleksandra Jasielska 2019 “Investigating Interpreters’ Empathy: Are Emotions in Simultaneous Interpreting Contagious?” Target 31 (1): 2–24. Google Scholar logo with link to Google Scholar; Li, Wang, and Rasmussen 2022Li, Saihong, Yifang Wang, and Yubo Zhou Rasmussen 2022 “Studying Interpreters’ Stress in Crisis Communication: Evidence from Multimodal Technology of Eye-Tracking, Heart Rate and Galvanic Skin Response.” The Translator 28 (4): 468–488. Google Scholar logo with link to Google ScholarLi, Saihong, Yifang Wang, and Yubo Zhou Rasmussen 2022 “Studying Interpreters’ Stress in Crisis Communication: Evidence from Multimodal Technology of Eye-Tracking, Heart Rate and Galvanic Skin Response.” The Translator 28 (4): 468–488. Google Scholar logo with link to Google Scholar; Korpal, Jankowiak, and Kaczmarek 2026Korpal, Paweł, Katarzyna Jankowiak, and Łukasz D. Kaczmarek 2026 “Emotional Content Affects Interpreters: Insights from Psychophysiological, Behavioral, and Self-Report Measures.” Interpreting 28 (1): 1–29. Google Scholar logo with link to Google ScholarKorpal, Paweł, Katarzyna Jankowiak, and Łukasz D. Kaczmarek 2026 “Emotional Content Affects Interpreters: Insights from Psychophysiological, Behavioral, and Self-Report Measures.” Interpreting 28 (1): 1–29. Google Scholar logo with link to Google Scholar); and acoustic-prosodic parameters such as voice intensity, pitch, and fluency (Kumcu 2023Kumcu, Alper 2023 “Emotional Language Processing in Bilingualism: Subjective Affect and Prosodic Markers in Simultaneous Interpreting.” Çankaya University Journal of Humanities and Social Sciences 17 (2): 194–211. Google Scholar logo with link to Google ScholarKumcu, Alper 2023 “Emotional Language Processing in Bilingualism: Subjective Affect and Prosodic Markers in Simultaneous Interpreting.” Çankaya University Journal of Humanities and Social Sciences 17 (2): 194–211. Google Scholar logo with link to Google Scholar) or speech rhythm of interpreter output as a function of syllabic intervals (Rojo López et al. 2021Rojo López, Ana María, Ana Isabel Foulquié Rubio, Laura Espín López, and Francisco Martínez Sánchez 2021 “Analysis of Speech Rhythm and Heart Rate as Indicators of Stress on Student Interpreters.” Perspectives 29 (4): 591–607. Google Scholar logo with link to Google ScholarRojo López, Ana María, Ana Isabel Foulquié Rubio, Laura Espín López, and Francisco Martínez Sánchez 2021 “Analysis of Speech Rhythm and Heart Rate as Indicators of Stress on Student Interpreters.” Perspectives 29 (4): 591–607. Google Scholar logo with link to Google Scholar), to name a few.

Short consecutive, or dialogue interpreting, which is predominant in community and healthcare settings, entails distinct stressors and cognitive load patterns for interpreters. Recent studies frame cognitive load in dialogue interpreting as a multidimensional construct comprising linguistic processing, memory retention, interactional coordination, and affective engagement (Geiling et al. 2021Geiling, Angelika, Christine Knaevelsrud, Maria Böttche, and Nadine Stammel 2021 “Mental Health and Work Experiences of Interpreters in the Mental Health Care of Refugees: A Systematic Review.” Frontiers in Psychiatry 12: 710789. Google Scholar logo with link to Google ScholarGeiling, Angelika, Christine Knaevelsrud, Maria Böttche, and Nadine Stammel 2021 “Mental Health and Work Experiences of Interpreters in the Mental Health Care of Refugees: A Systematic Review.” Frontiers in Psychiatry 12: 710789. Google Scholar logo with link to Google Scholar; Mellinger 2022Mellinger, Christopher D. 2022 “Cognitive Behaviour during Consecutive Interpreting: Describing the Notetaking Process.” Translation & Interpreting 14 (2): 103–119. Google Scholar logo with link to Google ScholarMellinger, Christopher D. 2022 “Cognitive Behaviour during Consecutive Interpreting: Describing the Notetaking Process.” Translation & Interpreting 14 (2): 103–119. Google Scholar logo with link to Google Scholar). Tiselius and Englund Dimitrova (2023)Tiselius, Elisabet, and Birgitta Englund Dimitrova 2023 “Testing the Working Memory Capacity of Dialogue Interpreters.” Across Languages and Cultures 24 (2): 163–180. Google Scholar logo with link to Google ScholarTiselius, Elisabet, and Birgitta Englund Dimitrova 2023 “Testing the Working Memory Capacity of Dialogue Interpreters.” Across Languages and Cultures 24 (2): 163–180. Google Scholar logo with link to Google Scholar were among the first to examine working memory and executive control in dialogue interpreting. Comparing experienced and novice dialogue interpreters across cognitive tasks and with a group of simultaneous interpreters, they found no effects of experience but apparent mode-specific differences. Their findings suggest that dialogue and simultaneous interpreting (SI) rely on different patterns of working memory allocation. Adler (2023)Adler, Aleksandra 2023 Cognitive Load in Dialogue Interpreting: Experience and Directionality. PhD diss. Stockholm University. http://​urn​.kb​.se​/resolve​?urn​=urn:nbn:se:su:diva​-220765Adler, Aleksandra 2023 Cognitive Load in Dialogue Interpreting: Experience and Directionality. PhD diss. Stockholm University. http://​urn​.kb​.se​/resolve​?urn​=urn:nbn:se:su:diva​-220765 demonstrated that interaction dynamics and interpersonal factors are important contributors to cognitive load in dialogue interpreting, reframing cognitive load as a socio-cognitive process. In another study, Adler (2024) 2024 “Effects of Experience and Directionality on Cognitive Load in Dialogue Interpreting.” Translation, Cognition & Behavior 7 (2): 187–208. Google Scholar logo with link to Google Scholar 2024 “Effects of Experience and Directionality on Cognitive Load in Dialogue Interpreting.” Translation, Cognition & Behavior 7 (2): 187–208. Google Scholar logo with link to Google Scholar demonstrated that interpreters experience greater cognitive load when working in their weaker language (L2) and that less experienced interpreters generally exhibit greater cognitive effort than their more experienced colleagues. Similarly, Yan et al. (2024)Yan, Hao, Yi Zhang, Yanqin Feng, et al. 2024 “Assessing Mental Demand in Consecutive Interpreting: Insights from an fNIRS Study.” Acta Psychologica 243: 104132. Google Scholar logo with link to Google ScholarYan, Hao, Yi Zhang, Yanqin Feng, et al. 2024 “Assessing Mental Demand in Consecutive Interpreting: Insights from an fNIRS Study.” Acta Psychologica 243: 104132. Google Scholar logo with link to Google Scholar investigated the impact of various mental workload components on the quality of CI, using neuroimaging (fNIRS), as well as subjective measures, and demonstrated that mental demand was the most significant factor influencing interpretation quality, and that its increase led to a negative effect on performance.

There are now several studies in the literature involving GSR and acoustic measures. Korpal and Jasielska (2019)Korpal, Paweł, and Aleksandra Jasielska 2019 “Investigating Interpreters’ Empathy: Are Emotions in Simultaneous Interpreting Contagious?” Target 31 (1): 2–24. Google Scholar logo with link to Google ScholarKorpal, Paweł, and Aleksandra Jasielska 2019 “Investigating Interpreters’ Empathy: Are Emotions in Simultaneous Interpreting Contagious?” Target 31 (1): 2–24. Google Scholar logo with link to Google Scholar examined SI from Polish (L1) to English (L2) in emotionally negative and neutral speech, measuring skin conductance (SC) and subjective emotional states using the PANAS scale. They found that interpreting negative-valence speeches consistently heightened interpreters’ physiological arousal, as shown by increased SC, and elicited negative emotional states. This pattern suggests the presence of automatic empathic resonance, in which interpreters spontaneously share the speaker’s emotions, and of emotional mimicry, in which interpreters unconsciously mirror the speaker’s affective tone. Complementing this, Rojo López et al. (2021)Rojo López, Ana María, Ana Isabel Foulquié Rubio, Laura Espín López, and Francisco Martínez Sánchez 2021 “Analysis of Speech Rhythm and Heart Rate as Indicators of Stress on Student Interpreters.” Perspectives 29 (4): 591–607. Google Scholar logo with link to Google ScholarRojo López, Ana María, Ana Isabel Foulquié Rubio, Laura Espín López, and Francisco Martínez Sánchez 2021 “Analysis of Speech Rhythm and Heart Rate as Indicators of Stress on Student Interpreters.” Perspectives 29 (4): 591–607. Google Scholar logo with link to Google Scholar demonstrated that phonological variables, specifically the rhythmic pattern of the interpreter’s speech, measured by the time intervals between successive syllable onsets (syllabic intervals), correlate with self-reported state anxiety (rather than trait anxiety) during CI, particularly from L1 to L2, reinforcing the notion that acoustic parameters are sensitive indicators of emotional stress. Similarly, Kumcu (2023)Kumcu, Alper 2023 “Emotional Language Processing in Bilingualism: Subjective Affect and Prosodic Markers in Simultaneous Interpreting.” Çankaya University Journal of Humanities and Social Sciences 17 (2): 194–211. Google Scholar logo with link to Google ScholarKumcu, Alper 2023 “Emotional Language Processing in Bilingualism: Subjective Affect and Prosodic Markers in Simultaneous Interpreting.” Çankaya University Journal of Humanities and Social Sciences 17 (2): 194–211. Google Scholar logo with link to Google Scholar investigated affective responses during interpreting from L1 (Turkish) to L2 (English) and found that interpreting emotionally negative speech elicited increased self-reported negative affect, whereas interpreting emotionally positive speech did not elicit changes in positive affect. Voice parameters revealed that voice intensity was generally indicative of cognitive load, while pitch and fluency were more sensitive to variations in emotional valence. These findings align with Scherer’s (1986)Scherer, Klaus R. 1986 “Voice, Stress, and Emotion.” In Dynamics of Stress, edited by Mortimer H. Appley and Richard Trumbull, 157–179. New York: Plenum Press. Google Scholar logo with link to Google ScholarScherer, Klaus R. 1986 “Voice, Stress, and Emotion.” In Dynamics of Stress, edited by Mortimer H. Appley and Richard Trumbull, 157–179. New York: Plenum Press. Google Scholar logo with link to Google Scholar observations on the vocal-auditory channel’s efficacy in reflecting emotional and motivational states, supported by ample evidence that vocal features are reliable markers of stress (Bachorowski and Owren 1995Bachorowski, Jo-Anne, and Michael J. Owren 1995 “Vocal Expression of Emotion: Acoustic Properties of Speech Are Associated with Emotional Intensity and Context.” Psychological Science 6 (4): 219–224. Google Scholar logo with link to Google ScholarBachorowski, Jo-Anne, and Michael J. Owren 1995 “Vocal Expression of Emotion: Acoustic Properties of Speech Are Associated with Emotional Intensity and Context.” Psychological Science 6 (4): 219–224. Google Scholar logo with link to Google Scholar; Pisanski, Groyecka-Bernard, and Sorokowski 2021Pisanski, Katarzyna, Agata Groyecka-Bernard, and Piotr Sorokowski 2021 “Human Voice Pitch Measures Are Robust across a Variety of Speech Recordings: Methodological and Theoretical Implications.” Biology Letters 17 (9): 20210356. Google Scholar logo with link to Google ScholarPisanski, Katarzyna, Agata Groyecka-Bernard, and Piotr Sorokowski 2021 “Human Voice Pitch Measures Are Robust across a Variety of Speech Recordings: Methodological and Theoretical Implications.” Biology Letters 17 (9): 20210356. Google Scholar logo with link to Google Scholar). Extending this research, Li, Wang, and Rasmussen (2022)Li, Saihong, Yifang Wang, and Yubo Zhou Rasmussen 2022 “Studying Interpreters’ Stress in Crisis Communication: Evidence from Multimodal Technology of Eye-Tracking, Heart Rate and Galvanic Skin Response.” The Translator 28 (4): 468–488. Google Scholar logo with link to Google ScholarLi, Saihong, Yifang Wang, and Yubo Zhou Rasmussen 2022 “Studying Interpreters’ Stress in Crisis Communication: Evidence from Multimodal Technology of Eye-Tracking, Heart Rate and Galvanic Skin Response.” The Translator 28 (4): 468–488. Google Scholar logo with link to Google Scholar utilised multimodal technology, namely eye-tracking, heart rate, and GSR, to examine interpreter stress in crisis communication scenarios involving healthcare interpreting during the COVID-19 pandemic. They observed significantly higher levels of physiological stress and cognitive effort during crisis interpreting than in everyday medical contexts. Furthermore, physiological stress and cognitive load adversely affected their performance. SC levels suggested differences between two different working conditions separated by induced stress (i.e., live TV broadcast and healthcare interpreting) (see also Kurz 2002Kurz, Ingrid 2002 “Physiological Stress Responses during Media and Conference Interpreting.” In Interpreting in the 21st Century: Challenges and Opportunities, edited by Giuliana Garzone and Maurizio Viezzi, 195–202. Amsterdam: John Benjamins. Google Scholar logo with link to Google ScholarKurz, Ingrid 2002 “Physiological Stress Responses during Media and Conference Interpreting.” In Interpreting in the 21st Century: Challenges and Opportunities, edited by Giuliana Garzone and Maurizio Viezzi, 195–202. Amsterdam: John Benjamins. Google Scholar logo with link to Google Scholar), but not between experts and novices (see also Kurz 2003 2003 “Physiological Stress during Simultaneous Interpreting: A Comparison of Experts and Novices.” The Interpreters’ Newsletter 12: 51–67.Google Scholar logo with link to Google Scholar 2003 “Physiological Stress during Simultaneous Interpreting: A Comparison of Experts and Novices.” The Interpreters’ Newsletter 12: 51–67.Google Scholar logo with link to Google Scholar).

Taken together, the abovementioned studies have shown that the stress experienced by interpreters frequently originates from heightened emotional and/or cognitive demands inherent to the task itself. These demands might include linguistic/textual complexity (Kuang and Zheng 2023Kuang, Huolingxiao, and Binghan Zheng 2023 “Note-Taking Effort in Video Remote Interpreting: Effects of Source Speech Difficulty and Interpreter Work Experience.” Perspectives 31 (4): 724–744. Google Scholar logo with link to Google ScholarKuang, Huolingxiao, and Binghan Zheng 2023 “Note-Taking Effort in Video Remote Interpreting: Effects of Source Speech Difficulty and Interpreter Work Experience.” Perspectives 31 (4): 724–744. Google Scholar logo with link to Google Scholar), fast speech and strong accents (Amato and Mack 2011Amato, Amalia, and Gabriele Mack 2011 “Interpreting the Oscar Night on Italian TV: An Interpreters’ Nightmare?” The Interpreters’ Newsletter 16: 37–60.Google Scholar logo with link to Google ScholarAmato, Amalia, and Gabriele Mack 2011 “Interpreting the Oscar Night on Italian TV: An Interpreters’ Nightmare?” The Interpreters’ Newsletter 16: 37–60.Google Scholar logo with link to Google Scholar; Korpal 2016Korpal, Paweł 2016 “Interpreting as a Stressful Activity: Physiological Measures of Stress in Simultaneous Interpreting.” Poznan Studies in Contemporary Linguistics 52 (2): 297–316. Google Scholar logo with link to Google ScholarKorpal, Paweł 2016 “Interpreting as a Stressful Activity: Physiological Measures of Stress in Simultaneous Interpreting.” Poznan Studies in Contemporary Linguistics 52 (2): 297–316. Google Scholar logo with link to Google Scholar), prolonged turns (Klonowicz 1990Klonowicz, Tatiana 1990 “A Psychophysiological Assessment of Simultaneous Interpreting: The Interaction of Individual Differences and Mental Workload.” Polish Psychological Bulletin 21: 37–48.Google Scholar logo with link to Google ScholarKlonowicz, Tatiana 1990 “A Psychophysiological Assessment of Simultaneous Interpreting: The Interaction of Individual Differences and Mental Workload.” Polish Psychological Bulletin 21: 37–48.Google Scholar logo with link to Google Scholar; Moser-Mercer, Künzli, and Korac 1998Moser-Mercer, Barbara, Alexander Künzli, and Marina Korac 1998 “Prolonged Turns in Interpreting: Effects on Quality, Physiological and Psychological Stress (Pilot Study).” Interpreting 3 (1): 47–64. Google Scholar logo with link to Google ScholarMoser-Mercer, Barbara, Alexander Künzli, and Marina Korac 1998 “Prolonged Turns in Interpreting: Effects on Quality, Physiological and Psychological Stress (Pilot Study).” Interpreting 3 (1): 47–64. Google Scholar logo with link to Google Scholar), time pressure (Gumul 2021Gumul, Ewa 2021 “Reporting Stress in Simultaneous Interpreting: The Analysis of Trainee Interpreters’ Retrospective Reports and Outputs.” Onomázein NE8: 16–42.Google Scholar logo with link to Google ScholarGumul, Ewa 2021 “Reporting Stress in Simultaneous Interpreting: The Analysis of Trainee Interpreters’ Retrospective Reports and Outputs.” Onomázein NE8: 16–42.Google Scholar logo with link to Google Scholar), challenging techno-social work environments (Kurz 2002Kurz, Ingrid 2002 “Physiological Stress Responses during Media and Conference Interpreting.” In Interpreting in the 21st Century: Challenges and Opportunities, edited by Giuliana Garzone and Maurizio Viezzi, 195–202. Amsterdam: John Benjamins. Google Scholar logo with link to Google ScholarKurz, Ingrid 2002 “Physiological Stress Responses during Media and Conference Interpreting.” In Interpreting in the 21st Century: Challenges and Opportunities, edited by Giuliana Garzone and Maurizio Viezzi, 195–202. Amsterdam: John Benjamins. Google Scholar logo with link to Google Scholar; Mackintosh 2003Mackintosh, Jennifer 2003 “The AIIC Workload Study.” Forum 1 (2): 189–214. Google Scholar logo with link to Google ScholarMackintosh, Jennifer 2003 “The AIIC Workload Study.” Forum 1 (2): 189–214. Google Scholar logo with link to Google Scholar), and dealing with role ambiguity (Hsieh 2006Hsieh, Elaine 2006 “Conflicts in How Interpreters Manage Their Roles in Provider–Patient Interactions.” Social Science & Medicine 62 (3): 721–730. Google Scholar logo with link to Google ScholarHsieh, Elaine 2006 “Conflicts in How Interpreters Manage Their Roles in Provider–Patient Interactions.” Social Science & Medicine 62 (3): 721–730. Google Scholar logo with link to Google Scholar). Although it has received comparatively less attention than the previously mentioned stressors, emotionally charged and potentially precarious communicative contexts involving emotion-laden words, expressions, and semantic-pragmatic structures can be a source of stress in interpreting, especially in mental healthcare settings (see Leanza, Pointurier, and Duchesne 2025Leanza, Yvan, Sophie Pointurier, and Laurie Duchesne 2025 “Interpreting Is an Emotional Effort: Public Service Interpreters’ Coping Strategies in Emotionally Charged Situations.” Interpreting and Society 5 (1): 3–26. Google Scholar logo with link to Google ScholarLeanza, Yvan, Sophie Pointurier, and Laurie Duchesne 2025 “Interpreting Is an Emotional Effort: Public Service Interpreters’ Coping Strategies in Emotionally Charged Situations.” Interpreting and Society 5 (1): 3–26. Google Scholar logo with link to Google Scholar).

The brief review of studies above suggests that stress experienced during interpreting has both emotional and cognitive sources and leaves traces on several biomarkers. Furthermore, individuals differ considerably in how they perceive and evaluate experiences as stressful, underscoring the need to identify individual predictors of stress. It is notoriously difficult to disentangle the psychophysiological effects of cognitive and emotional constructs and to determine their individual contributions. That said, there is some evidence suggesting that cognitive and emotional loads might have different effects on interpreting. For example, Kumcu (2023)Kumcu, Alper 2023 “Emotional Language Processing in Bilingualism: Subjective Affect and Prosodic Markers in Simultaneous Interpreting.” Çankaya University Journal of Humanities and Social Sciences 17 (2): 194–211. Google Scholar logo with link to Google ScholarKumcu, Alper 2023 “Emotional Language Processing in Bilingualism: Subjective Affect and Prosodic Markers in Simultaneous Interpreting.” Çankaya University Journal of Humanities and Social Sciences 17 (2): 194–211. Google Scholar logo with link to Google Scholar found that voice intensity generally reflects changes in cognitive effort during SI. However, the fundamental frequency (F 0 ) and pause ratio are more responsive to shifts in source speech valence: lower valence results in lower F 0 and longer pauses.

Against this background, the current study aims to distinguish the effects of emotional and cognitive load in mental healthcare interpreting by examining three research questions:

  1. RQ1: Is cognitive or emotional load the more dominant factor in the interpreter’s stress?

  2. RQ2: Which psycho-affective measures (skin conductance, voice parameters, self-reports) are related to which type of load in CI?

  3. RQ3: Does stress affect CI performance, and if so, is cognitive or emotional load more detrimental?

Interpreter-mediated psychotherapy sessions were chosen as the context because communication in mental healthcare settings may inherently carry some potential for emotional content. Similarly, CI was selected because it is the primary modality for interpreting in mental healthcare.

We operationalised cognitive load and emotional load as two distinct variables. Cognitive load, in this study, refers to the mental effort required to process, retain, and produce verbal input in the target language under time and memory constraints, and is intrinsic to CI. It is influenced by linguistic complexity, semantic density, lexical richness, and lexical density, as well as the unpredictability of the verbal material to be interpreted. Emotional load refers to the emotional demands placed on the interpreter by their exposure to negative or distressing content. It also refers to the possibility that participants may become emotionally engaged with what the speaker describes. This psychological experience is quite common among interpreters working in community interpreting settings, and processing it requires effort (Crezee, Jülich, and Hayward 2013Crezee, Ineke, Shirley Jülich, and Maria Hayward 2013 “Issues for Interpreters and Professionals Working in Refugee Settings.” Journal of Applied Linguistics and Professional Practice 8 (3): 253–273. Google Scholar logo with link to Google ScholarCrezee, Ineke, Shirley Jülich, and Maria Hayward 2013 “Issues for Interpreters and Professionals Working in Refugee Settings.” Journal of Applied Linguistics and Professional Practice 8 (3): 253–273. Google Scholar logo with link to Google Scholar; Geiling et al. 2022 2022 “Psychological Distress, Exhaustion, and Work-Related Correlates among Interpreters Working in Refugee Care: Results of a Nationwide Online Survey in Germany.” European Journal of Psychotraumatology 13 (1): 2046954. Google Scholar logo with link to Google Scholar 2022 “Psychological Distress, Exhaustion, and Work-Related Correlates among Interpreters Working in Refugee Care: Results of a Nationwide Online Survey in Germany.” European Journal of Psychotraumatology 13 (1): 2046954. Google Scholar logo with link to Google Scholar). Despite our reliance on such operational definitions, measuring stress and emotion in interpreting remains highly challenging. Cognitive load in interpreting is a multidimensional concept with various definitions and measurement approaches (Seeber 2011Seeber, Kilian G. 2011 “Cognitive Load in Simultaneous Interpreting: Existing Theories — New Models.” Interpreting 13 (2): 176–204. Google Scholar logo with link to Google ScholarSeeber, Kilian G. 2011 “Cognitive Load in Simultaneous Interpreting: Existing Theories — New Models.” Interpreting 13 (2): 176–204. Google Scholar logo with link to Google Scholar; Chen 2017Chen, Sijia 2017 “The Construct of Cognitive Load in Interpreting and Its Measurement.” Perspectives 25 (4): 640–657. Google Scholar logo with link to Google ScholarChen, Sijia 2017 “The Construct of Cognitive Load in Interpreting and Its Measurement.” Perspectives 25 (4): 640–657. Google Scholar logo with link to Google Scholar; Gieshoff and Hunziker Heeb 2023Gieshoff, Anne Catherine, and Andrea Hunziker Heeb 2023 “Cognitive Load and Cognitive Effort: Probing the Psychological Reality of a Conceptual Difference.” Translation, Cognition & Behavior 6 (1): 3–28. Google Scholar logo with link to Google ScholarGieshoff, Anne Catherine, and Andrea Hunziker Heeb 2023 “Cognitive Load and Cognitive Effort: Probing the Psychological Reality of a Conceptual Difference.” Translation, Cognition & Behavior 6 (1): 3–28. Google Scholar logo with link to Google Scholar). Similarly, stress resulting from emotional processing during interpreting can be observed not only in physiological arousal but also in subjective psychological experience (Korpal 2021 2021 “Stress and Emotion in Conference Interpreting.” In The Routledge Handbook of Conference Interpreting, edited by Michaela Albl-Mikasa and Elisabet Tiselius, 401–413. London: Routledge. Google Scholar logo with link to Google Scholar 2021 “Stress and Emotion in Conference Interpreting.” In The Routledge Handbook of Conference Interpreting, edited by Michaela Albl-Mikasa and Elisabet Tiselius, 401–413. London: Routledge. Google Scholar logo with link to Google Scholar). For this reason, this study adopted a holistic, multimethod approach to detect signs of stress during interpreting, using both physiological (GSR and speech output) and self-report measures (PANAS, STAI, and NASA-TLX), as detailed below. We also measured CI accuracy as a cumulative end product using an item-based analysis to examine whether stress affects performance.

2.Methods

2.1Participants

The study was conducted with twenty third- and fourth-year student interpreters from two universities in Türkiye (10 participants from Hacettepe University and 10 participants from Bilkent University) (11 males; M age = 23.65, SD = 2.78, range: 22–35). All participants were native Turkish speakers, having learned and spoken Turkish from birth and using it as their primary language, and had English as a second language. For purposes of interpreting, the participants’ A language was Turkish, and their B language was English. In the pre-test questionnaire all participants indicated having completed undergraduate-level CI courses, during which they had practised CI, the main task in the current study. The mean cumulative grade point average (CGPA) of the sample was 3.42 out of 4.00 (SD = 0.40, range = 2.26–3.94). Participants reported no voice, speech, or hearing problems, and no history of neurological disorders. The handedness test, conducted to define the dominant hand for GSR measurement, revealed that seventeen participants were right-handed, two were ambidextrous, and one was left-handed. The State-Trait Anxiety Inventory (STAI) (Spielberger et al. 1983Spielberger, Charles D., Richard L. Gorsuch, Robert E. Lushene, Paul R. Vagg, and Gerard A. Jacobs 1983 Manual for the State-Trait Anxiety Inventory (Form Y1–Y2). Palo Alto: Consulting Psychologists Press.Google Scholar logo with link to Google ScholarSpielberger, Charles D., Richard L. Gorsuch, Robert E. Lushene, Paul R. Vagg, and Gerard A. Jacobs 1983 Manual for the State-Trait Anxiety Inventory (Form Y1–Y2). Palo Alto: Consulting Psychologists Press.Google Scholar logo with link to Google Scholar), which has a total score range of 20–80, revealed that the mean trait anxiety score for the sample was 52.1 (SD = 10.21, range = 33–65). This level of trait anxiety implies that participants were generally prone to experiencing higher state anxiety in demanding situations such as interpreting tasks. All participants were fully informed of the procedure details and provided informed consent. The Ethics Committee of Bilkent University approved the study (approval number: 2024_01_25_01). Post-test debriefing revealed that all participants were initially unaware of the study’s purpose.

2.2Stimuli

The experiment was designed in a 2×2 factorial fashion (i.e., 2 independent variables with 2 levels). Emotional load and cognitive load were identified as the two independent variables, each with a higher and lower level. Thus, four Turkish speeches, corresponding to higher and lower levels of emotional and cognitive load, were developed. The speeches, as per the conditions, were labelled as (1) low emotional load + low cognitive load (LELC), (2) low emotional load + high cognitive load (LEHC), (3) high emotional load + low cognitive load (HELC), and (4) high emotional load + high cognitive load (HEHC). The descriptive characteristics of the speeches are presented in Table 1.

Table 1.Descriptive characteristics of the source speeches, including temporal and textual measures, readability scores, NLP-based profiling metrics, and speech-averaged word norms for four subdimensions of emotionality and concreteness. Standard deviations are shown in parentheses where available. MTLD = measure of textual lexical diversity
Measure High emotion
High cognition		 High emotion
Low cognition		 Low emotion
High cognition		 Low emotion
Low cognition
Duration (sec.) 140.13 179.12 133.51 168.83
Word count 350 350 350 350
Token count 414 447 397 439
Type count 260 289 258 272
Sentence count 37 81 37 85
Mean syllables per word 2.10 2.02 2.08 2.10
Mean sentence length 9.46 4.33 9.46 4.12
Ateşman 60.04 (Medium) 78.67 (Easy) 58.43 (Medium) 77.60 (Easy)
Çetinkaya-Uzun 54.21 52.59 53.17 51.36
Surprisal 5.98 6.60 6.09 6.01
Lexical density 0.83 0.82 0.79 0.80
Semantic density 0.65 0.64 0.72 0.72
Type/token ratio 0.70 0.77 0.68 0.70
MTLD 218.35 294.86 190.06 181.40
Arousal 5.75 (0.96) 5.36 (0.99) 4.93 (0.77) 4.89 (0.98)
Valence 4.94 (1.64) 4.89 (1.57) 5.70 (1.2) 5.48 (1.14)
Happiness 43.25 (29.83) 35.64 (26.06) 49.26 (19.55) 42.99 (21.43)
Sadness 36.69 (22.75) 35.37 (21.08) 23.61 (13.02) 24.16 (13.63)
Concreteness 51.39 (24.48) 60.99 (24.53) 59.69 (24.6) 55.6 (23.76)

All four speeches were fictional first-person narratives constructed for use in a simulated therapeutic setting in mental healthcare. The LELC speech featured a person describing their meditation practice, which their therapist had recommended to improve concentration. The LEHC speech was narrated by a person updating their therapist on their mild insomnia and recent progress. The HELC speech was produced by a person in recovery from heroin use, reflecting on a difficult period that involved partner-directed aggression and the challenges of withdrawal. Finally, the HEHC speech was delivered by a person recalling their mother’s violent behaviour towards them during their childhood as a traumatic experience. The speeches were recorded with a camcorder in a soundproof room by a male native Turkish speaker. Video files, complete transcripts, and their English translations are available in the study’s online data repository: https://​osf​.io​/2j749.

2.3Norming

We adopted a multi-pronged approach, and the speeches were normed and controlled through a three-step procedure: (1) text-based norming, (2) human ratings (experts and participants), and (3) NLP-based profiling. Results are presented in Table 1 and Table 2.

2.3.1Text-based norming

The transcripts were first analysed for difficulty and complexity as a function of readability using two Turkish adaptations of the Flesch Reading Ease formula (Ateşman 1997Ateşman, Ender 1997 “Türkçede okunabilirliğin ölçülmesi [Measuring readability in Turkish].” Dil Dergisi 58: 71–74.Google Scholar logo with link to Google ScholarAteşman, Ender 1997 “Türkçede okunabilirliğin ölçülmesi [Measuring readability in Turkish].” Dil Dergisi 58: 71–74.Google Scholar logo with link to Google Scholar; Çetinkaya and Uzun 2018Çetinkaya, Gökhan, and Leyla Uzun 2018 “Türkçe ders kitaplarındaki metinlerin okunabilirlik özellikleri [Readability features of texts in Turkish textbooks].” In Türkçe ders kitabı çözümlemeleri [Analyses of Turkish textbooks], edited by Hakan Ülper, 141–156. Ankara: Pegem Akademi.Google Scholar logo with link to Google ScholarÇetinkaya, Gökhan, and Leyla Uzun 2018 “Türkçe ders kitaplarındaki metinlerin okunabilirlik özellikleri [Readability features of texts in Turkish textbooks].” In Türkçe ders kitabı çözümlemeleri [Analyses of Turkish textbooks], edited by Hakan Ülper, 141–156. Ankara: Pegem Akademi.Google Scholar logo with link to Google Scholar), which are based on syllable count and sentence length. Both indices showed consistent differences between the speeches with higher and lower cognitive load: speeches with higher cognitive load (i.e., HEHC and LEHC) fell into the medium difficulty category (50–69), while speeches with lower cognitive load (i.e., HELC and LELC) fell into the easy category (70–89) according to Ateşman’s Readability Index.

Second, the speech transcripts were tokenised, and words that overlapped with those in the Turkish Emotional Word Norms (Kapucu et al. 2021Kapucu, Aycan, Aslı Kılıç, Yıldız Özkılıç, and Bengisu Sarıbaz 2021 “Turkish Emotional Word Norms for Arousal, Valence, and Discrete Emotion Categories.” Psychological Reports 124 (1): 188–209. Google Scholar logo with link to Google ScholarKapucu, Aycan, Aslı Kılıç, Yıldız Özkılıç, and Bengisu Sarıbaz 2021 “Turkish Emotional Word Norms for Arousal, Valence, and Discrete Emotion Categories.” Psychological Reports 124 (1): 188–209. Google Scholar logo with link to Google Scholar) were identified. The mean values of four subdimensions of emotionality (i.e., arousal, valence, happiness, and sadness) were calculated for each speech based on the overlapping words. Finally, the speeches were statistically compared against each other on these subdimensions. The results showed that, on average, words in the high emotionality speeches (i.e., HELC and HEHC) had higher arousal and sadness ratings (both ps < .0001) and lower valence ratings (p = .001) than those in the low emotionality speeches (i.e., LELC and LEHC). Furthermore, there were no significant differences between speeches in the same emotionality group (i.e., between HELC and HEHC or between LELC and LEHC speeches) in any of the subdimensions of emotionality (p > .05; see also Table 1).

2.3.2Human ratings

Speech transcripts were rated by fourteen language experts on two separate scales: emotionality (1 = not emotional, 5 = very emotional) and valence (1 = very negative, 5 = very positive) (see Table 2). As expected, results showed significant differences between high (emotionality M = 4.11, SD = 0.92; valence M = 1.54, SD = 0.58) and low emotionality speeches (emotionality M = 2.32, SD = 1.12; valence M = 3.46, SD = 0.74), both in emotionality (t(51.9) = 6.52, p < .0001) and valence (t(50.81) = 10.84, p < .0001), but not between speeches within the same emotionality group (i.e., between HELC and HEHC and between LELC and LEHC speeches; p > .05).

The speeches were also internally validated by all study participants (n = 20) at the end of each CI task for emotionality, valence, interpreting difficulty, imageability (vividness), familiarity, speed and self-performance on a scale from 1 (not emotional, very negative, very easy, not imageable, not familiar, very slow, and very unsatisfactory) to 5 (very emotional, very positive, very difficult, highly imageable, very familiar, very fast, and very satisfactory). Consistent with the expert norming, results revealed significant differences between high (emotionality M = 4.15, SD = 0.95; valence M = 1.38, SD = 0.67) and low emotional speeches (emotionality M = 1.93, SD = 0.94; valence M = 3.42, SD = 0.84) both in emotionality (t(77.998) = 10.51, p < .0001) and valence (t(74.07) = 12.05, p < .0001). Further, there was no difference between high (M = 2.8, SD = 1.09) and low cognitive load speeches (M = 2.72, SD = 1.01) in interpreting difficulty (t(77.57) = 0.32, p = .75). There were also no significant differences between the speeches in terms of familiarity or self-performance (p > .05). There was only a small but significant difference (p = .01) in imageability between the HELC (M = 4.35, SD = 0.81) and the LEHC speech (M = 3.3, SD = 1.17).

Table 2.Expert (n = 14) and participant (n = 20) norming results for the source speeches (means and SDs)
Variable High emotion
High cognition		 High emotion
Low cognition		 Low emotion
High cognition		 Low emotion
Low cognition
Expert norming
Emotionality  4.43 (0.65)  3.79 (1.05)  2.14 (1.03)  2.5 (1.22)
Valence 1.5 (0.52)  1.57 (0.65)  3.21 (0.7)   3.71 (0.73)
Participant norming
Emotionality  4.25 (1.07)  4.05 (0.83)  1.85 (0.93) 2 (0.97)
Valence  1.35 (0.67) 1.4 (0.68)  3.35 (0.75)   3.5 (0.95)
Difficulty 2.9 (1.21) 2.9 (1.02) 2.7 (0.98) 2.55 (1)
Imageability 3.9 (1.21)  4.35 (0.81) 3.3 (1.17)   3.75 (1.07)
Familiarity  2.95 (1.39)  3.05 (1.28)  3.75 (0.91)  3.9 (1.17)
Speed  2.95 (0.83) 2.6 (0.82)  2.75 (0.91)  2.8 (0.77)
Self-performance  2.95 (1.05)  3.35 (1.18) 3.1 (0.85)  3.2 (0.89)

2.3.3NLP-based profiling

Our design required systematic control over cognitive versus emotional load in naturalistic CI speech. The traditional text descriptors reported in Section 2.3.1 (e.g., readability formulas, word and sentence count, length) help approximate difficulty but may overlook core determinants of processing effort in interpreting (Puerini and Muñoz Martín 2025Puerini, Sara, and Ricardo Muñoz Martín 2025 “Beyond Readability Formulas: Text Profiling in Cognitive Translation & Interpreting Studies.” Paper presented at Fifth International Conference on Translation, Interpreting and Cognition, Kristiansand, Norway. June 2–4.Puerini, Sara, and Ricardo Muñoz Martín 2025 “Beyond Readability Formulas: Text Profiling in Cognitive Translation & Interpreting Studies.” Paper presented at Fifth International Conference on Translation, Interpreting and Cognition, Kristiansand, Norway. June 2–4.), notably predictability, semantic drift, and lexical content density. To address this, we complemented readability indices and human ratings with NLP-based profiling that directly targets the constructs theoretically linked to cognitive load and incremental comprehension in CI. The following NLP/ML (machine learning) metrics were used: surprisal, semantic density, lexical density, and lexical richness.

Surprisal indexes contextual unpredictability at the word-by-word level. Surprisal was computed as a function of the average negative log-likelihood per token for each text using a pre-trained large Turkish GPT-2 language model (ytu-ce-cosmos/turkish-gpt2-large) via the Hugging Face library (Kesgin et al. 2024Kesgin, H. Toprak, M. Kaan Yuce, Eren Dogan, M. Egemen Uzun, Atahan Uz, H. Emre Seyrek, Ahmed Zeer, and M. Fatih Amasyali 2024 “Introducing cosmosGPT: Monolingual Training for Turkish Language Models.” Preprint, arXiv. https://​arxiv​.org​/abs​/2404​.17336 Kesgin, H. Toprak, M. Kaan Yuce, Eren Dogan, M. Egemen Uzun, Atahan Uz, H. Emre Seyrek, Ahmed Zeer, and M. Fatih Amasyali 2024 “Introducing cosmosGPT: Monolingual Training for Turkish Language Models.” Preprint, arXiv. https://​arxiv​.org​/abs​/2404​.17336 ). Higher surprisal is associated with increased moment-to-moment integration costs, which is theoretically congruent with CI demands on segmentation, retention, and reformulation.

Semantic density captures conceptual dispersion within a speech. Greater dispersion implies more topic shifts or lower coherence, increasing the need for chunking and planning. To calculate semantic density, sentence embeddings were generated using a multilingual sentence transformer model (sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2) via the Hugging Face library (Reimers and Gurevych 2019Reimers, Nils, and Iryna Gurevych 2019 “Sentence-BERT: Sentence Embeddings Using Siamese BERT-Networks.” In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), edited by Kentaro Inui, Jing Jiang, Vincent Ng, and Xiaojun Wan, 3982–3992. Stroudsburg, PA: Association for Computational Linguistics. Google Scholar logo with link to Google ScholarReimers, Nils, and Iryna Gurevych 2019 “Sentence-BERT: Sentence Embeddings Using Siamese BERT-Networks.” In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), edited by Kentaro Inui, Jing Jiang, Vincent Ng, and Xiaojun Wan, 3982–3992. Stroudsburg, PA: Association for Computational Linguistics. Google Scholar logo with link to Google Scholar). Mean pairwise cosine distance across sentence embeddings was then calculated to derive the mean semantic spread.

Lexical density refers to the ratio of content words (nouns, verbs, adjectives, and adverbs) to the overall word count in a speech. Thus, lexical density gauges the information per word, a classic proxy of propositional load relevant to note-free CI. Another lexical index, lexical richness, was computed using the type-token ratio (TTR, the ratio of unique to total words) and the measure of textual lexical diversity (MTLD, a robust lexical diversity metric that is resistant to text length). These measures reflect type diversity and style variability beyond length effects. That is, higher richness can tax selection, reformulation, and monitoring, even when readability is similar.

The results of NLP-based profiling are reported in Table 1. Results showed that, as intended, both types of high cognition speeches were cognitively demanding, albeit in different ways. The HEHC speech had the highest lexical density and the lowest surprisal score, while the LEHC speech had the joint-highest semantic density, despite a lower TTR. Conversely, the HELC speech exhibited the highest lexical richness, as measured by both TTR and MTLD, as well as the highest surprisal value, indicating emotional spontaneity and lexical variety without consistent structural complexity. This lexical richness is assumed to reflect stylistic and emotional expressiveness rather than the systematic linguistic or syntactic complexity that typically imposes a greater cognitive load during interpreting. The LELC speech showed an average level of lexical density and semantic variation, as well as the lowest MTLD. This is consistent with the prediction that simpler and more predictable speech is less cognitively and emotionally demanding.

2.4Procedure

The main task of the study was CI from L1 (Turkish) to L2 (English). This directionality was chosen so that participants would encounter the source speeches in L1, in which emotional language tends to elicit stronger responses than in an L2 (Harris, Ayçiçeği, and Gleason 2003Harris, Catherine L., Ayşe Ayçiçeği, and Jean Berko Gleason 2003 “Taboo Words and Reprimands Elicit Greater Autonomic Reactivity in a First Language than in a Second Language.” Applied Psycholinguistics 24 (4): 561–579. Google Scholar logo with link to Google ScholarHarris, Catherine L., Ayşe Ayçiçeği, and Jean Berko Gleason 2003 “Taboo Words and Reprimands Elicit Greater Autonomic Reactivity in a First Language than in a Second Language.” Applied Psycholinguistics 24 (4): 561–579. Google Scholar logo with link to Google Scholar; Caldwell-Harris et al. 2011Caldwell-Harris, Catherine L., Jimmy Tong, Winvy Lung, and Sinlan Poo 2011 “Physiological Reactivity to Emotional Phrases in Mandarin–English Bilinguals.” International Journal of Bilingualism 15 (3): 329–352. Google Scholar logo with link to Google ScholarCaldwell-Harris, Catherine L., Jimmy Tong, Winvy Lung, and Sinlan Poo 2011 “Physiological Reactivity to Emotional Phrases in Mandarin–English Bilinguals.” International Journal of Bilingualism 15 (3): 329–352. Google Scholar logo with link to Google Scholar), consistent with the relevant literature on interpreting (e.g., Korpal and Jasielska 2019Korpal, Paweł, and Aleksandra Jasielska 2019 “Investigating Interpreters’ Empathy: Are Emotions in Simultaneous Interpreting Contagious?” Target 31 (1): 2–24. Google Scholar logo with link to Google ScholarKorpal, Paweł, and Aleksandra Jasielska 2019 “Investigating Interpreters’ Empathy: Are Emotions in Simultaneous Interpreting Contagious?” Target 31 (1): 2–24. Google Scholar logo with link to Google Scholar). Presenting the source speeches in Turkish was therefore intended to increase the likelihood of eliciting emotional responses during source-speech processing, allowing us to examine how this emotional input affected interpreting into English. As described below, the study was composed of three consecutive phases: (1) the baseline task, (2) CI with GSR and speech recording, and (3) post-task testing.

The baseline task was carried out one week before the interpreting task. Participants were asked to describe a black-and-white drawing of a street. Following brief instructions that served as a minimal, uniform preparation period, participants began speaking as soon as the image appeared on the screen. In doing so, participants produced baseline SC and voice data for neutral speech, which was then compared with the CI output. Baseline values were not used to standardise or normalise the physiological and vocal data. Instead, the baseline task was modelled as an additional speech-type category in the analyses, allowing us to compare electrodermal activity and voice parameters during a neutral speech-production task with electrodermal activity during listening and electrodermal activity and voice parameters during interpreting. Conceptually, we opted for a speech-based baseline rather than a resting baseline because interpreting is fundamentally a speech-production task. We assumed that comparing interpreting-related arousal with a neutral speech task provided a reference point for assessing interpreting-induced stress. Accordingly, we attempted to reduce the risk of bias due to the anticipation of a stressful task. At this stage, participants were also asked eleven questions about their demographic characteristics (age and gender), academic background (university, year of study, CGPA, courses taken, and interpreting experience), and medical history (handedness, medication use, alcohol consumption, and any neurological or psychiatric diagnoses). Finally, they completed two self-report measures of their emotional states at baseline. The first was the Positive and Negative Affect Schedule (PANAS) (Watson, Clark, and Tellegen 1988Watson, David, Lee Anna Clark, and Auke Tellegen 1988 “Development and Validation of Brief Measures of Positive and Negative Affect: The PANAS Scales.” Journal of Personality and Social Psychology 54 (6): 1063–1070. Google Scholar logo with link to Google ScholarWatson, David, Lee Anna Clark, and Auke Tellegen 1988 “Development and Validation of Brief Measures of Positive and Negative Affect: The PANAS Scales.” Journal of Personality and Social Psychology 54 (6): 1063–1070. Google Scholar logo with link to Google Scholar), an assessment tool to assess mood and emotional states, consisting of two ten-item (i.e., adjective) subscales measuring positive affect (e.g., enthusiasm, alertness) and negative affect (e.g., distress, anxiety) using a five-point Likert scale (scores ranging from 10 to 50 per subscale). The second was the State-Trait Anxiety Inventory (STAI) (Spielberger et al. 1983Spielberger, Charles D., Richard L. Gorsuch, Robert E. Lushene, Paul R. Vagg, and Gerard A. Jacobs 1983 Manual for the State-Trait Anxiety Inventory (Form Y1–Y2). Palo Alto: Consulting Psychologists Press.Google Scholar logo with link to Google ScholarSpielberger, Charles D., Richard L. Gorsuch, Robert E. Lushene, Paul R. Vagg, and Gerard A. Jacobs 1983 Manual for the State-Trait Anxiety Inventory (Form Y1–Y2). Palo Alto: Consulting Psychologists Press.Google Scholar logo with link to Google Scholar), an assessment tool for measuring anxiety levels, consisting of two twenty-item subscales measuring state anxiety (temporary, situational) and trait anxiety (long-term, dispositional) using a four-point Likert scale (scores range from 20 to 80 per subscale).

In the second phase, CI with GSR and speech recording was carried out. Participants were asked to interpret four speeches, categorised according to their emotionality and cognitive difficulty as described in Section 2.3 (i.e., LELC, LEHC, HELC, HEHC), consecutively from L1 (Turkish) to L2 (English). Participants received no prior information about the speeches or their terminological background and performed CI without note-taking while watching and listening to the source speech videos. The speeches were presented in a Latin-squared counterbalanced order. Electrodermal activity was measured using the ProComp Infiniti system (Thought Technology Ltd, Canada). The GSR sensor was placed on the palmar surface of the dominant hand, specifically on the distal phalanges of the index and middle fingers, in accordance with standard electrode placement protocols. The data were sampled at 256 Hz to provide a high-resolution record of electrodermal activity. The choice of a 256 Hz sampling rate is consistent with previous studies using the ProComp Infiniti system, thereby ensuring consistent and comparable data collection (e.g., Yang, Lin, and Wu 2025Yang, Po-Chou, I-Mei Lin, and Da-Wei Wu 2025 “Effects of Heart Rate Variability Biofeedback on Enhancing Self-Efficacy, Quality of Life and Six-Minute Walking Test in Patients with Chronic Obstructive Pulmonary Disease.” Applied Psychophysiology and Biofeedback 50: 403–416. Google Scholar logo with link to Google ScholarYang, Po-Chou, I-Mei Lin, and Da-Wei Wu 2025 “Effects of Heart Rate Variability Biofeedback on Enhancing Self-Efficacy, Quality of Life and Six-Minute Walking Test in Patients with Chronic Obstructive Pulmonary Disease.” Applied Psychophysiology and Biofeedback 50: 403–416. Google Scholar logo with link to Google Scholar). Skin temperature was also measured as a control, but was not included in the analysis. Each recording session began with a two-minute warm-up and calibration period. The interpreters’ speech output was recorded using an external microphone and Audacity version 3.7.1 (Audacity Team) for the analysis of voice parameters and accuracy.

The third phase consisted of post-task testing. Participants were asked to complete the PANAS (see Korpal and Jasielska 2019Korpal, Paweł, and Aleksandra Jasielska 2019 “Investigating Interpreters’ Empathy: Are Emotions in Simultaneous Interpreting Contagious?” Target 31 (1): 2–24. Google Scholar logo with link to Google ScholarKorpal, Paweł, and Aleksandra Jasielska 2019 “Investigating Interpreters’ Empathy: Are Emotions in Simultaneous Interpreting Contagious?” Target 31 (1): 2–24. Google Scholar logo with link to Google Scholar, for a similar procedure) and the STAI (see Rojo López et al. 2021Rojo López, Ana María, and Ana Isabel Foulquié Rubio 2024 “Interpreting, Affect, and Emotion.” In The Routledge Handbook of Interpreting and Cognition, edited by Christopher D. Mellinger, 307–323. London: Routledge. Google Scholar logo with link to Google ScholarRojo López, Ana María, and Ana Isabel Foulquié Rubio 2024 “Interpreting, Affect, and Emotion.” In The Routledge Handbook of Interpreting and Cognition, edited by Christopher D. Mellinger, 307–323. London: Routledge. Google Scholar logo with link to Google Scholar, for a similar procedure) immediately after each of the four CI tasks, as they did at the baseline task. They were also asked to complete the NASA Task Load Index (NASA-TLX) (Hart and Staveland 1988Hart, Sandra G., and Lowell E. Staveland 1988 “Development of NASA-TLX (Task Load Index): Results of Empirical and Theoretical Research.” In Human Mental Workload, edited by Peter A. Hancock and Najmedin Meshkati, 139–183. Amsterdam: Elsevier.Google Scholar logo with link to Google ScholarHart, Sandra G., and Lowell E. Staveland 1988 “Development of NASA-TLX (Task Load Index): Results of Empirical and Theoretical Research.” In Human Mental Workload, edited by Peter A. Hancock and Najmedin Meshkati, 139–183. Amsterdam: Elsevier.Google Scholar logo with link to Google Scholar) (see Sun and Shreve 2014Sun, Sanjun, and Gregory M. Shreve 2014 “Measuring Translation Difficulty: An Empirical Study.” Target 26 (1): 98–127. Google Scholar logo with link to Google ScholarSun, Sanjun, and Gregory M. Shreve 2014 “Measuring Translation Difficulty: An Empirical Study.” Target 26 (1): 98–127. Google Scholar logo with link to Google Scholar; Yan et al. 2024Yan, Hao, Yi Zhang, Yanqin Feng, et al. 2024 “Assessing Mental Demand in Consecutive Interpreting: Insights from an fNIRS Study.” Acta Psychologica 243: 104132. Google Scholar logo with link to Google ScholarYan, Hao, Yi Zhang, Yanqin Feng, et al. 2024 “Assessing Mental Demand in Consecutive Interpreting: Insights from an fNIRS Study.” Acta Psychologica 243: 104132. Google Scholar logo with link to Google Scholar, for similar procedures). The NASA-TLX consists of six workload dimensions (mental, physical, and temporal demands, as well as performance, effort, and frustration), each rated on a 21-point scale. Finally, after each CI task, participants completed a questionnaire in which they rated the perceived emotionality, valence, interpreting difficulty, imageability (vividness), familiarity, and speed of the source speech, as well as their own CI performance. Importantly, they were also asked whether they had had a similar personal experience with the events described in the speech.

In summary, participants in the present study were given a neutral baseline task (describing a black-and-white picture) one week before the main experiment. Then, they consecutively interpreted four short speeches, varying in cognitive and emotional load, from Turkish into English. We collected data using two physiological measures (SC and speech output), three self-report measures (PANAS, STAI, and NASA-TLX), and a performance measure based on accuracy.

3.Results

3.1Data analysis approach

3.1.1Skin conductance

Raw GSR data were exported using the BioGraph Infiniti software (Thought Technology Ltd., Canada). The main indicator was skin conductance (SC), expressed in microsiemens (µS). The data were pre-processed initially by applying a low-pass filter at 1 Hz to remove high-frequency noise and artefacts. Within the recording, two distinct epochs of interest were analysed: a listening phase, immediately succeeded by an interpreting phase, which was considered a continuation of the cognitive processes initiated during listening. These epochs were extracted using stimulus-onset markers. Artefact-detection criteria were used to ensure data reliability, and segments containing excessive movement artefacts or missing values were removed. All analyses adhered to the established guidelines for processing electrodermal activity (Boucsein et al. 2012Boucsein, Wolfram, Don C. Fowles, Sverre Grimnes, Gershon Ben-Shakhar, Walton T. Roth, Michael E. Dawson, and Diane L. Filion 2012 “Publication Recommendations for Electrodermal Measurements.” Psychophysiology 49 (8): 1017–1034. Google Scholar logo with link to Google ScholarBoucsein, Wolfram, Don C. Fowles, Sverre Grimnes, Gershon Ben-Shakhar, Walton T. Roth, Michael E. Dawson, and Diane L. Filion 2012 “Publication Recommendations for Electrodermal Measurements.” Psychophysiology 49 (8): 1017–1034. Google Scholar logo with link to Google Scholar) and methodological conventions in the literature (Rojo López et al. 2025Rojo López, Ana María, Marina Ramos Caro, Daria Patalas, and Aleksandra Jasielska 2025 “Psychophysiological Methods.” In Research Methods in Cognitive Translation and Interpreting Studies, edited by Ana María Rojo López and Ricardo Muñoz Martín, 256–278. Amsterdam: John Benjamins. Google Scholar logo with link to Google ScholarRojo López, Ana María, Marina Ramos Caro, Daria Patalas, and Aleksandra Jasielska 2025 “Psychophysiological Methods.” In Research Methods in Cognitive Translation and Interpreting Studies, edited by Ana María Rojo López and Ricardo Muñoz Martín, 256–278. Amsterdam: John Benjamins. Google Scholar logo with link to Google Scholar). The main assumption underlying electrodermal activity is that higher SC values indicate heightened physiological arousal experienced by a participant (Benedek and Kaernbach 2010Benedek, Mathias, and Christian Kaernbach 2010 “A Continuous Measure of Phasic Electrodermal Activity.” Journal of Neuroscience Methods 190 (1): 80–91. Google Scholar logo with link to Google ScholarBenedek, Mathias, and Christian Kaernbach 2010 “A Continuous Measure of Phasic Electrodermal Activity.” Journal of Neuroscience Methods 190 (1): 80–91. Google Scholar logo with link to Google Scholar).

3.1.2Auditory output

Participants’ auditory output was analysed using Praat version 6.3.03 (Boersma and Weenink 2022Boersma, Paul, and David Weenink 2022 Praat: Doing Phonetics by Computer. Version 6.3.03. Amsterdam. http://​www​.praat​.org/Boersma, Paul, and David Weenink 2022 Praat: Doing Phonetics by Computer. Version 6.3.03. Amsterdam. http://​www​.praat​.org/). We used the psychological stress indicators proposed by Sondhi et al. (2015)Sondhi, Savita, Munna Khan, Ritu Vijay, Ashok K. Salhan, and Satish Chouhan 2015 “Acoustic Analysis of Speech under Stress.” International Journal of Bioinformatics Research and Applications 11 (5): 417–432. Google Scholar logo with link to Google ScholarSondhi, Savita, Munna Khan, Ritu Vijay, Ashok K. Salhan, and Satish Chouhan 2015 “Acoustic Analysis of Speech under Stress.” International Journal of Bioinformatics Research and Applications 11 (5): 417–432. Google Scholar logo with link to Google Scholar to analyse auditory output: mean voice intensity (dB), mean pitch as a function of F 0 (Hz), and pitch variation as a function of F 0 SD. The F 0 range was set at 75–300 Hz for males and 75–500 Hz for females, as per convention. As the first step, we removed all silent pauses so that the sound files contained only speech segments. Then, we exported aggregated measures; that is, a single value was exported for each voice indicator per participant. Output durations were recorded but not further analysed due to their dependency on source-speech length.

3.1.3Statistical models

The data were analysed using linear mixed-effects models (LMEMs) and generalised additive mixed-effects models (GAMMs) to identify predictors of self-reports, SC, mean voice intensity (dB), mean F 0 , F 0 variation, and accuracy. To disentangle cognitive and emotional load, the data were modelled in two complementary ways: (1) combined-factor models, which included speech type as the fixed effect with either five levels (baseline, LELC, LEHC, HELC, and HEHC) or four levels (when the baseline task was dropped), thus combining low and high cognitive and emotional load through speech types; and (2) factorial models, which included two orthogonal factors (cognitive and emotional load) with two levels each (low and high), along with their main and interaction effects, which were tested separately. The phase was also modelled as a fixed effect with three levels (baseline, listening, and interpreting) for SC. Fixed effects were entered as predictors, and participants as random intercepts. The contributions of the fixed effects were assessed by comparing full models with reduced or null models (Winter 2013Winter, Bodo 2013 “Linear Models and Linear Mixed Effects Models in R with Linguistic Applications.” Preprint, arXiv. http://​arxiv​.org​/abs​/1308​.5499Winter, Bodo 2013 “Linear Models and Linear Mixed Effects Models in R with Linguistic Applications.” Preprint, arXiv. http://​arxiv​.org​/abs​/1308​.5499).

We used GAMMs to analyse SC exclusively because models based on mean SC values cannot capture the temporal dynamics that unfold during the listening and interpreting phases. GAMMs incorporate a smooth time term that allows SC to vary non-linearly across the duration of each task, which is essential given that SC is a high-frequency, continuous signal that changes over time. In contrast, LMEMs were applied to voice parameters because Praat provides segment-level averages rather than frame-by-frame trajectories, making LMEMs the more suitable approach for these discrete acoustic measures.

LMEMs were preferred over ANOVA to avoid the Language-as-Fixed-Effect Fallacy (Clark 1973Clark, Herbert H. 1973 “The Language-as-Fixed-Effect Fallacy: A Critique of Language Statistics in Psychological Research.” Journal of Verbal Learning and Verbal Behavior 12 (4): 335–359. Google Scholar logo with link to Google ScholarClark, Herbert H. 1973 “The Language-as-Fixed-Effect Fallacy: A Critique of Language Statistics in Psychological Research.” Journal of Verbal Learning and Verbal Behavior 12 (4): 335–359. Google Scholar logo with link to Google Scholar) and to account for random participant variability (Baayen, Davidson, and Bates 2008Baayen, R. Harald, Douglas J. Davidson, and Douglas M. Bates 2008 “Mixed-Effects Modeling with Crossed Random Effects for Subjects and Items.” Journal of Memory and Language 59 (4): 390–412. Google Scholar logo with link to Google ScholarBaayen, R. Harald, Douglas J. Davidson, and Douglas M. Bates 2008 “Mixed-Effects Modeling with Crossed Random Effects for Subjects and Items.” Journal of Memory and Language 59 (4): 390–412. Google Scholar logo with link to Google Scholar). By modelling participants as random effects, LMEMs accommodated individual differences in intercepts, yielding more generalisable and statistically robust inferences. The data were analysed and visualised in R (R Core Team 2025R Core Team 2025 R: A Language and Environment for Statistical Computing. Vienna: R Foundation for Statistical Computing. https://​www​.r​-project​.org/R Core Team 2025 R: A Language and Environment for Statistical Computing. Vienna: R Foundation for Statistical Computing. https://​www​.r​-project​.org/) using the lme4 (Bates et al. 2015Bates, Douglas M., Martin Mächler, Ben Bolker, and Steve Walker 2015 “Fitting Linear Mixed-Effects Models Using lme4.” Journal of Statistical Software 67 (1): 1–48. Google Scholar logo with link to Google ScholarBates, Douglas M., Martin Mächler, Ben Bolker, and Steve Walker 2015 “Fitting Linear Mixed-Effects Models Using lme4.” Journal of Statistical Software 67 (1): 1–48. Google Scholar logo with link to Google Scholar), lmerTest (Kuznetsova, Brockhoff, and Christensen 2017Kuznetsova, Alexandra, Per B. Brockhoff, and Rune Haubo Bojesen Christensen 2017 “lmerTest Package: Tests in Linear Mixed Effects Models.” Journal of Statistical Software 82 (13): 1–26. Google Scholar logo with link to Google ScholarKuznetsova, Alexandra, Per B. Brockhoff, and Rune Haubo Bojesen Christensen 2017 “lmerTest Package: Tests in Linear Mixed Effects Models.” Journal of Statistical Software 82 (13): 1–26. Google Scholar logo with link to Google Scholar), and mgcv (Wood 2011Wood, Simon N. 2011 “Fast Stable Restricted Maximum Likelihood and Marginal Likelihood Estimation of Semiparametric Generalized Linear Models.” Journal of the Royal Statistical Society (B) 73 (1): 3–36. Google Scholar logo with link to Google ScholarWood, Simon N. 2011 “Fast Stable Restricted Maximum Likelihood and Marginal Likelihood Estimation of Semiparametric Generalized Linear Models.” Journal of the Royal Statistical Society (B) 73 (1): 3–36. Google Scholar logo with link to Google Scholar) packages (see the online data repository at https://​osf​.io​/2j749 for the analysis codes, and raw and processed data).

3.2Self-report measures

Combined-factor mixed-effects models were fitted with speech type (baseline task and four speeches) as a fixed effect, participants as a random effect, and positive affect score on the PANAS as the target variable. Results showed that speech type significantly improved the null positive affect model (χ 2(4) = 14.09, p = .007). Tukey post-hoc tests showed that participants reported significantly lower positive affect after interpreting all four speech types relative to the baseline task. The estimated coefficients indicate that the decrease was largest for the HEHC speech (M = 28.55, SD = 7.33); B = −4.55, z = −3.61, p = .0003, followed by the LELC (M = 29.25, SD = 8.25); B = −3.85, z = −3.05, p = .002, LEHC (M = 29.9, SD = 7.57); B = −3.20, z = −2.54, p = .01, and HELC speeches (M = 30.6, SD = 8.74); B = −2.50, z = −1.98, p = .05. These coefficients derive from the underlying mixed-effects model and reflect the direction and magnitude of each contrast tested in the Tukey procedure. Factorial models with cognitive and emotional load as fixed effects revealed that neither cognitive load nor emotional load nor their interaction significantly improved the null positive affect model (p > .05).

Similar models were fitted with the PANAS negative affect score as the target variable. Speech type improved the null negative affect model (χ 2(4) = 11.33, p = .02). Post-hoc tests showed that participants reported significantly higher negative affect scores after interpreting HEHC speech (M = 18.9, SD = 9.48) than after LELC speech (M = 14.95, SD = 3.66); B = 3.95, z = 2.79, p = .005, LEHC speech (M = 15.5, SD = 4.84); B = 3.4, z = 2.4, p = .02, and after the baseline task (M = 14.8, SD = 4.38); B = 4.1, z = 2.89, p = .004 (see Figure 1). Factorial models with cognitive and emotional load as fixed effects showed that only emotional load improved the null negative affect model (χ 2(1) = 6.90, p = .01). As expected, interpreting emotional speeches increased self-reported negative affect (B = 2.80, t = 2.71, p = .01). Further, negative affect showed a significant cognitive-emotional load interaction (B = 3.95, t = 2.73, p = .008), with the strongest negative affect reported under simultaneous high cognitive and emotional load.

The same models were fitted for state anxiety measured with the STAI. Because trait anxiety reliably influences state anxiety levels, we included participants’ trait anxiety scores as a covariate in all anxiety models. As expected, trait anxiety was a strong predictor of state anxiety in every model (ps < .0001). In contrast, speech type, cognitive load, emotional load, or their interaction did not account for additional variance in state anxiety once trait anxiety was controlled for (p > .05).

Similarly, there was no significant difference between the speeches on any of the six workload dimensions of the NASA-TLX (p > .05). Factorial models revealed that neither cognitive load nor emotional load nor their interaction significantly improved the null NASA-TLX models (p > .05). Most of the self-report measures were intercorrelated in the expected directions (see the online data repository at https://​osf​.io​/2j749 for the correlation matrix).

Models were also fitted with speech characteristics, as determined by participant ratings, as fixed effects, and self-report measures as target variables. Results showed that interpreting more emotional speeches (B = 1.04, t = 2.62, p = .01), more negative speeches (B = −1.42, t = −3.45, p = .001), and more imageable and thus, vivid speeches (B = 1.24, t = 2.18, p = .03) resulted in higher self-reported negative affect. Interpreting more difficult speeches resulted in both lower positive affect (B = −1.11, t = −2.07, p = .04) and higher negative affect (B = 1.52, t = 2.45, p = .02). Likewise, interpreting speeches associated with higher self-perceived performance resulted in both higher positive affect (B = 2.19, t = 4.44, p < .0001), and lower negative affect (B = −1.42, t = −2.27, p = .03). Finally, interpreting more familiar speeches led to higher positive affect (B = 1.25, t = 2.96, p = .004), but did not predict negative affect (p = .69). Emotionality, valence, and imageability did not predict positive affect. Self-perceived speed did not predict either positive or negative affect (p > .05). Results are presented in Table 3.

Figure 1.Half violin plots showing differences in positive and negative affect scores as per PANAS across the speech conditions
Figure 1.

3.3Skin conductance

3.3.1Difference between the phases

Mixed-effects models were fitted with phase as a fixed effect, participants as a random effect, and mean SC as the target variable. Results showed that the phase significantly improved the null SC model (χ 2(2) = 8.40, p = .015), revealing a gradual increase in SC from baseline through the listening and interpreting phases (see Figure 2). The mean SC during the interpreting phase (M = 15.35, SD = 6.06) was significantly higher than the mean SC during the baseline phase (M = 11.43, SD = 5.26); B = 3.93, z = 3.06, p = .002, but not significantly higher than the mean SC during the listening phase (M = 13.38, SD = 4.98); B = 1.98, z = 1.54, p = .12. Another model was fitted with the speech type as a fixed effect, participants as a random effect and the mean SC as the target variable. Speech type did not improve the null model of mean SC (χ 2(3) = 3.08, p = .38), indicating that interpreting different speeches did not alter the mean SC. Similarly, there was no significant difference in mean SC between the speech types during the listening phase (χ 2(3) = 0.81, p = .85). Lastly, factorial models for the listening and interpreting phases were fitted with cognitive and emotional load as fixed effects. Factorial models revealed that neither cognitive load nor emotional load nor their interaction significantly improved the null mean GSR models for the listening or interpreting phases (p > .05). Further analyses revealed that no speech characteristics, as measured by participant ratings, were associated with the mean SC during either interpreting or listening (p > .05). Results are presented in Table 3.

Figure 2.Half violin plots showing mean SC across the speech conditions during listening and interpreting
Figure 2.

3.3.2Time course of skin conductance

We fitted GAMMs to analyse the effects of speech types, emotional load, and cognitive load on interpreters’ SC across time. The first model was fitted for the interpreting phase, including speech type as a fixed effect. Participants showed a significant decrease in SC during the interpretation of LELC speech (B = −1.33, t = −115.97, p < .0001) and HELC speech (B = −1.00, t = −82.28, p < .0001) relative to the interpretation of HEHC speech (the reference level in the model). No significant difference in SC was observed during the interpretation of LEHC speech (B = −0.01, t = −0.74, p = .46) compared to HEHC speech. The smooth term for the time was significant (F = 10,493, p < .0001), indicating a substantial temporal effect on SC. The combined-factor model suggested that the HEHC condition produced the highest overall SC. Both LELC and HELC conditions yielded reliably lower SC, while LEHC did not differ significantly from HEHC. This suggests that cognitive load played a central role, whereas emotional load alone was insufficient to elevate SC to the highest levels in the interpreting phase.

A factorial model was fitted with emotional and cognitive load as fixed effects to investigate their additive influence and to complement the combined-factor model. The model showed that both factors significantly increased SC. A high emotional load was associated with a modest increase in SC relative to a low emotional load (B = 0.17, t = 20.99, p < .0001). In contrast, high cognitive load produced a substantially larger increase in SC (B = 1.16, t = 131.22, p < .0001). The smooth term for time remained highly significant, indicating non-linear fluctuations in SC over the course of the task (F = 10,885, p < .0001). Taken together, similar to speech-type models, factorial models revealed that, although both emotional and cognitive load increased arousal, cognitive load exerted a markedly stronger influence on SC than emotional load during the interpreting phase (see Figure 3). The factorial GAMM also revealed a negative, significant interaction between emotional and cognitive load (B = −0.32, t = −19.26, p < .0001). This negative interaction indicates that the combined effect of high emotional and high cognitive load on SC was sub-additive. That is, each factor independently increased SC, yet their joint presence did not produce the sum of their separate effects.

The exact same analysis procedure was followed for the listening phase. Relative to the reference level in the model (HEHC), SC was significantly lower in the HELC (B = −0.62, t = −71.52, p < .0001), LELC (B = −0.46, t = −52.06, p < .0001), and LEHC (B = −0.09, t = −10.44, p < .0001) conditions. These effects indicate that the HEHC condition consistently elicited the highest SC, whereas the LELC and HELC conditions elicited reduced physiological arousal.

To disentangle the contributions of emotional and cognitive load in the listening phase, we fitted a GAMM with additive effects of each factor. Both predictors were significant: SC was slightly higher in the low emotional load condition than in the high emotional load condition (B = 0.05, t = 7.39, p < .0001). SC was substantially lower in the low cognitive load condition versus the high cognitive load condition (B = −0.50, t = −78.60, p < .0001). These coefficients indicate that a high cognitive load is associated with a significant increase in SC. In contrast, emotional load has a much smaller effect in the listening phase, as in the interpreting phase (see Figure 3). Lastly, the factorial model, which included the interaction between emotional and cognitive load, replicated the main patterns while revealing a significant interaction term (B = 0.26, t = 21.23, p < .0001). The interaction indicates that the difference between high and low emotional load depends on cognitive load. When cognitive load is low, emotional load has a stronger impact on SC than when cognitive load is high. The smooth term for time was a significant main effect in all models (ps < .0001).

We visualised SC trends over time during the listening and interpreting phases across speech conditions to clarify the effects of each load (see Figure 3). The curves show GAMM-smoothed mean SC (µS) based on participant-level data. To reduce oversmoothing, the data were binned into 100-ms intervals within participant × speech type × phase, and the mean SC in each bin was used for smoothing. The dashed vertical line marks the transition between the listening and interpreting phases, defined as the midpoint between the final bin labelled ‘listening’ and the first bin labelled ‘interpreting’. Because the original recordings were continuous, a small temporal overlap between the end of listening and the onset of interpreting may occur in some speeches, reflecting natural variation in the timing of participants’ interpretation onset. The dashed line, therefore, represents an approximate boundary for visualisation purposes.

Figure 3.Plot showing SC trends over time during the listening and interpreting phases for each speech condition. Shaded ribbons around each curve indicate 95% confidence intervals
Figure 3.

3.4Voice parameters

3.4.1Voice intensity

Combined-factor mixed-effects models were fitted with speech type as a fixed effect, participants as a random effect, and mean voice intensity (in dB) as the target variable. The results showed that the speech type significantly improved the null voice intensity model (χ 2(4) = 12.12, p = .02). Post-hoc analysis showed that the mean voice intensity was significantly lower when interpreting LEHC speech (M = 64.19, SD = 4.75); B = −3.65, z = −3.14, p = .002, LELC speech (M = 64.48, SD = 6.15); B = −3.36, z = −2.88, p = .004, HEHC speech (M = 64.61, SD = 4.96); B = −3.23, z = −2.78, p = .006, and HELC speech (M = 65.07, SD = 4.89); B = −2.77, z = −2.38, p = .02 as compared to the baseline task (M = 67.84, SD = 5.15) (see Figure 4 and Table 3). Speech type did not improve the null voice intensity model when the baseline task condition was dropped (χ 2(3) = 0.73, p = .87), indicating no difference between speech types. Cognitive and emotional load, or their interactions, as fixed effects in factorial models did not significantly improve the null voice intensity models (p > .05). Models with speech characteristics as fixed effects showed that only self-perceived speed predicted voice intensity. As speed increased, mean voice intensity decreased (B = −1.37, t = −2.17, p = .03).

3.4.2Mean fundamental frequency

Mixed-effects models were fitted, with speech type as a fixed effect, participants as a random effect, and mean F 0 as the target variable. The speech type significantly improved the null F 0 model (χ 2(4) = 11.40, p = .02). Mean F 0 was lower when interpreting HELC speech (M = 145.91, SD = 41.13) as compared to the baseline task (M = 152.92, SD = 41.95); B = −7, z = −3.23, p = .001, LELC speech (M = 151.64, SD = 39.06); B = −5.72, z = −2.64, p = .008, and LEHC speech (M = 150.89, SD = 42.98); B = −4.97, z = −2.29, p = .02 (see Figure 4 and Table 3). When the baseline task was dropped, the speech type also improved the null F 0 model (χ 2(3) = 9.98, p = .02), with the same differences observed in HELC speech as compared to LEHC and LELC speeches. Additional models with cognitive load and emotional load as fixed effects revealed that emotional load improved the null F 0 model (χ 2(1) = 6.34, p = .01), whereas cognitive load did not (p = .33). In general, participants produced a higher F 0 when interpreting low emotional speeches than high emotional speeches (B = 3.56, t = 2.59, p = .01). However, F 0 showed a significant effect of emotional load under low cognitive load (B = −5.72, t = −3.03, p = .004). Specifically, mean F 0 was lower in HELC speech than in LELC speech, suggesting that elevated emotional load alone reduced F 0 . Models with speech characteristics as fixed effects showed that higher familiarity with the content of the source speech was associated with higher mean F 0 (B = 2.41, t = 3.67, p = .0005).

3.4.3Fundamental frequency variation

Speech type did not improve the null model when the SD of the F 0 as a function of pitch variation was added as a target variable (χ 2(4) = 7.96, p = .09). That said, post-hoc analysis showed that pitch variation was lower during the interpretation of HELC speech (M = 28.99, SD = 13.18); B = −7.75, z = −2.81, p = .005 and LELC speech (M = 31.29, SD = 16.54); B = −5.45, z = −1.98, p = .05 as compared to the baseline task (M = 36.74, SD = 21.37). Factorial models revealed that neither cognitive load nor emotional load nor their interaction significantly improved the null F 0 variation models (p > .05). Models with speech characteristics as fixed effects revealed that lower difficulty predicted higher pitch variation (B = −2.20, t = −2.43, p = .02).

Figure 4.Half violin plots showing the mean voice intensity, mean pitch, and pitch variation across the speech conditions
Figure 4.

3.5Accuracy and source-target synchronisation

To answer the third research question, we assessed interpreters’ accuracy, defined as the extent to which they convey the semantic content of the source speech in their target language output. Accuracy was calculated using a unit-based analysis in which semantic units from the source speech were weighted according to their communicative importance, following principles similar to those proposed by Gieshoff and Albl-Mikasa (2024)Gieshoff, Anne Catherine, and Michaela Albl-Mikasa 2024 “Interpreting Accuracy Revisited: A Refined Approach to Interpreting Performance Analysis.” Perspectives 32 (2): 210–228. Google Scholar logo with link to Google ScholarGieshoff, Anne Catherine, and Michaela Albl-Mikasa 2024 “Interpreting Accuracy Revisited: A Refined Approach to Interpreting Performance Analysis.” Perspectives 32 (2): 210–228. Google Scholar logo with link to Google Scholar. These units were then evaluated using established accuracy criteria in the interpreting literature (Lee 2008Lee, Jieun 2008 “Rating Scales for Interpreting Performance Assessment.” The Interpreter and Translator Trainer 2 (2): 165–184. Google Scholar logo with link to Google ScholarLee, Jieun 2008 “Rating Scales for Interpreting Performance Assessment.” The Interpreter and Translator Trainer 2 (2): 165–184. Google Scholar logo with link to Google Scholar; Han 2018Han, Chao 2018 “Using Rating Scales to Assess Interpretation: Practices, Problems and Prospects.” Interpreting 20 (1): 63–101. Google Scholar logo with link to Google ScholarHan, Chao 2018 “Using Rating Scales to Assess Interpretation: Practices, Problems and Prospects.” Interpreting 20 (1): 63–101. Google Scholar logo with link to Google Scholar), and total rendition scores were computed for each speech. Combined-factor models showed no significant differences in CI accuracy across the four speech types (p > .05). Likewise, factorial models with cognitive and emotional load as fixed effects revealed that neither cognitive load nor emotional load nor their interaction significantly improved the null accuracy model (p > .05). Additional mixed-effects analyses showed that skin conductance, voice parameters, and self-report measures did not significantly predict CI accuracy (all ps > .05).

Finally, we explored source-target synchronisation in voice parameters to further explain the changes in voice parameters across speeches. Item-level correlations between the source and target speeches were used instead of mixed-effects modelling due to the small sample size (n = 4). As expected, given the small sample size, the correlations were not statistically significant. However, the coefficients suggested notable patterns between the source and target speech, including a strong positive correlation in mean pitch (r(2) = .75), pitch variation (r(2) = .54), as well as a negative correlation in voice intensity (r(2) = −.67).

Table 3.Self-reported scores from the PANAS, STAI, and NASA-TLX, together with mean skin conductance, voice parameters (means and SDs), and accuracy across the speech conditions
Measure Baseline High emotion
High cognition		 High emotion
Low cognition		 Low emotion
High cognition		 Low emotion
Low cognition
Positive affect 33.1 (6.12) 28.55 (7.33) 30.6 (8.74) 29.9 (7.57) 29.25 (8.25)
Negative affect 14.8 (4.38) 18.9 (9.48) 17.15 (6.47) 15.5 (4.84) 14.95 (3.66)
State anxiety 36.9 (10.57) 39.5 (13.87) 39.55 (13.48) 38.05 (10.18) 38.05 (10.11)
Mental demand 13.05 (4.73) 12.85 (4.49) 11.4 (5.59) 11.95 (5.29)
Physical demand 4.5 (4.49) 4.35 (3.3) 3.5 (3.02) 3.8 (2.76)
Temporal demand 9.65 (5.33) 9 (4.8) 8.65 (5.37) 8.65 (4.99)
Performance 10.3 (4.46) 8.1 (5) 9.05 (3.1) 8.8 (3.98)
Effort 13.8 (4.47) 13.8 (4.57) 13.35 (5.03) 12.4 (5.22)
Frustration 10.7 (6.7) 10.9 (6.64) 9.25 (6.1) 8.95 (5.49)
Baseline SC 11.43 (5.26)
Listening SC 13.57 (5.52) 13.2 (5.01) 13.58 (4.72) 13.16 (5.77)
Interpreting SC 15.53 (6.38) 15.46 (5.64) 15.76 (5.96) 14.66 (7.27)
Mean intensity (dB) 67.84 (5.15) 64.61 (4.96) 65.07 (4.89) 64.19 (4.75) 64.48 (6.15)
Mean F 0 (Hz) 152.92 (41.95) 149.49 (41.32) 145.91 (41.13) 150.89 (42.98) 151.64 (39.06)
F 0 variation (Hz) 36.74 (21.37) 32.07 (17.77) 28.99 (13.18) 32.58 (17.32) 31.29 (16.54)
Accuracy 41.29 (10.71) 39.92 (11.44) 38.33 (6.95) 38.17 (9.28)

4.Discussion

The current study aimed to identify the sources of stress interpreters experience when they interpret and to clarify the differential effects of emotional and cognitive load on stress. Specifically, we examined self-reported affect, anxiety, and mental demand, as well as skin responses and voice parameters during CI of short monologues from L1 (Turkish) to L2 (English). To the best of our knowledge, this is the first attempt to systematically disentangle emotional and cognitive load in interpreting by combining physiological, acoustic, and subjective measures. We also leveraged NLP/ML techniques to rigorously control stimulus complexity and emotionality within a factorial design. Our hybrid norming approach, which integrates readability indices, expert evaluations, participant ratings, and NLP-based profiling, offered three key advantages. First, we aimed to ensure comprehensive construct coverage by capturing both predictability and semantic drift in interpreting, dimensions that traditional readability measures alone cannot reflect. Second, we attempted to achieve scale invariance, as metrics such as MTLD minimise text-length bias and enable relatively fair comparisons across speeches of different lengths. Third, the tools we used provided model-based control tailored to the linguistic characteristics of Turkish via a Turkish language model and multilingual sentence embeddings, ensuring methodological appropriateness for our L1–L2 context.

The most important finding of the study was that physiological arousal during the interpreting process, as measured by skin conductance, was driven by cognitive factors rather than emotional ones. In other words, interpreting more difficult source speeches led to higher arousal, as the cognitive load corresponded to the difficulty of interpreting in the present study due to the way the source speeches were manipulated. There is ample evidence from the study to support this outcome.

First, the higher mean SC during CI tasks than during the baseline task, regardless of speech type, can be explained by the greater cognitive challenge posed by CI than by delivering an impromptu speech. Second, GAMM analysis of skin response over time revealed that, although both cognitive and emotional load triggered arousal during CI, cognitive load had a much more pronounced effect than emotional load. This is primarily indicated by the factorial model, which simultaneously assesses the additive effects of cognitive and emotional load on SC within the same model. Further evidence in this respect was the absence of a significant difference in arousal between the two cognitively demanding speeches that differed solely in terms of emotionality (i.e., HEHC and LEHC). Conversely, participants experienced the highest levels of arousal when interpreting the highly emotional, cognitively demanding speech (i.e., HEHC) compared to both lower cognitive load speeches (i.e., HELC and LELC), regardless of emotionality.

The plot showing changes in SC over time (Figure 3) further illustrates the predominance of cognitive load over emotional load. The time-course patterns indicate that both the listening and interpreting phases can be divided into two distinct stages, each comprising an early and a late half. The impact of cognitive load on arousal becomes particularly evident in the later stages of both tasks. In the initial stage of the interpreting phase, cognitive and emotional load jointly ordered the speeches by their effects on mean SC, with cognitive load exerting a more pronounced influence (i.e., HEHC, LEHC, HELC, and LELC from highest to lowest). In the final stage, however, the two high-cognition speeches (i.e., HEHC and LEHC) were clearly separated from the two lower cognitive load speeches (i.e., HELC and LELC), regardless of emotional load. While SC continued to rise toward the end of the interpreting phase for the high-cognition speeches, it declined for the lower cognitive load speeches.

Figure 3 visually suggests that HELC speech elicits higher SC than LELC speech, especially towards the end of the time course. Although the difference between LELC and HELC speeches was not statistically significant in the GAMM analysis, this does not contradict our main conclusion. We claim that, across both listening and interpreting phases, cognitive load produced far larger and more systematic increases in SC than emotional load, not that emotional load has no effect at all. Similarly, the plot suggests that the LEHC speech elicits a slightly higher skin response than the HEHC speech, also towards the end of the interpreting phase. Likewise, this LEHC–HEHC difference was not statistically significant in the GAMM analysis. One likely contributing factor is that the LEHC speech had a higher semantic density than the HEHC speech. Another explanation concerns imageability: participants rated LEHC as the least imageable (i.e., most abstract) of the four speeches. Given that note-taking was not allowed, interpreting a more abstract speech may have imposed a heavier burden on working memory, as mental imagery control is a key predictor of CI performance (Kumcu and Öztürk 2023Kumcu, Alper, and Asiye Öztürk 2023 “Visual Mental Imagery and Verbal Working Memory: Evidence from Consecutive Interpreting.” Journal of Cognitive Psychology 35 (5): 545–560. Google Scholar logo with link to Google ScholarKumcu, Alper, and Asiye Öztürk 2023 “Visual Mental Imagery and Verbal Working Memory: Evidence from Consecutive Interpreting.” Journal of Cognitive Psychology 35 (5): 545–560. Google Scholar logo with link to Google Scholar). The pattern has been observed toward the end because, as the interval between listening and production increases, the strain on working memory likely increases, requiring interpreters to update stored information while integrating incoming cues (Dong, Liu, and Cai 2018Dong, Yanping, Yuhua Liu, and Rendong Cai 2018 “How Does Consecutive Interpreting Training Influence Working Memory: A Longitudinal Study of Potential Links between the Two.” Frontiers in Psychology 9: 875. Google Scholar logo with link to Google ScholarDong, Yanping, Yuhua Liu, and Rendong Cai 2018 “How Does Consecutive Interpreting Training Influence Working Memory: A Longitudinal Study of Potential Links between the Two.” Frontiers in Psychology 9: 875. Google Scholar logo with link to Google Scholar). Fatigue toward the end of the interpreting phase may have further amplified this effect. In any case, these divergences should be interpreted cautiously, as they emerge in the final segments where fewer data points remain, potentially introducing visual fluctuations that are not supported by the model estimates.

At first glance, the effect of cognitive load on arousal may appear counterintuitive, given the intensity of the emotional content included in the high-emotion speeches (e.g., abuse, addiction, trauma). However, this result is both plausible and consistent with dual-task processing models (Pashler 1994Pashler, Harold 1994 “Dual-Task Interference in Simple Tasks: Data and Theory.” Psychological Bulletin 116 (2): 220–244. Google Scholar logo with link to Google ScholarPashler, Harold 1994 “Dual-Task Interference in Simple Tasks: Data and Theory.” Psychological Bulletin 116 (2): 220–244. Google Scholar logo with link to Google Scholar). One possible explanation is that, when emotional and cognitive demands coincide, the interpreter’s limited attentional and working memory resources are primarily allocated to managing cognitive demands, such as segmenting the input, reformulating the message, and resolving lexical or syntactic issues. In such cases, immediate emotional processing at the time of interpreting may become secondary, particularly in L2 output contexts where expressive resources are already limited (Schmaußer and Laborde 2023Schmaußer, Maximilian, and Sylvain Laborde 2023 “Tonic and Phasic Cardiac Vagal Activity Predict Cognitive-Affective Processing in an Emotional Stop-Signal Task.” International Journal of Psychophysiology 191: 9–18. Google Scholar logo with link to Google ScholarSchmaußer, Maximilian, and Sylvain Laborde 2023 “Tonic and Phasic Cardiac Vagal Activity Predict Cognitive-Affective Processing in an Emotional Stop-Signal Task.” International Journal of Psychophysiology 191: 9–18. Google Scholar logo with link to Google Scholar). Another plausible explanation is that interpreters may engage in emotional regulation strategies, either consciously or unconsciously (see Augustine and Hemenover 2009Augustine, Adam A., and Scott H. Hemenover 2009 “On the Relative Effectiveness of Affect Regulation Strategies: A Meta-Analysis.” Cognition and Emotion 23 (6): 1181–1220. Google Scholar logo with link to Google ScholarAugustine, Adam A., and Scott H. Hemenover 2009 “On the Relative Effectiveness of Affect Regulation Strategies: A Meta-Analysis.” Cognition and Emotion 23 (6): 1181–1220. Google Scholar logo with link to Google Scholar for a comprehensive review), when confronted with negative emotions. This could reduce the physiological reactivity associated with emotional load, enabling the cognitive demands of the task to take precedence. Participants’ ability to suppress emotional mimicry or disengage empathically could therefore reduce SC variability attributable to emotional load.

Notably, participants experienced arousal not only when interpreting difficult speeches, but also when listening to them. GAMM analysis revealed that listening to difficult-to-interpret speeches with high cognitive load elicited greater arousal than those with low cognitive load. The mean SC during the listening phase of the HEHC speech increased more than in the other three speeches, standing out clearly in the final stage. We argue that this demonstrates not only the effect of cognitive load, as discussed above, but also the strategic and purposeful nature of ‘listening for interpreting’, which involves memory encoding without note-taking, and how it differs from listening without a clear goal, that is, passive listening. The real-life-like first-person narratives in mental healthcare contexts used in the present study may have prompted anticipatory cognitive planning during the listening phase, especially in high-complexity conditions. This could explain why arousal was elevated even before the interpreting task began, suggesting that interpreters pre-load cognitively in anticipation of an upcoming difficult task. In simpler terms, because listening for CI involves planning and preparing for the subsequent interpreting phase, participants might have realised that the HEHC speech would be challenging to interpret, and this might have increased skin conductance in response. There is evidence that anticipatory emotional responses, such as anxiety, trigger changes in behavioural, autonomic, and neural response systems, preparing individuals for impending challenges, as evidenced by changes in pupil diameter, heart rate (Jercic, Sennersten, and Lindley 2017Jercic, Petar, Charlotte Sennersten, and Craig Lindley 2017 “The Effect of Cognitive Load on Physiological Arousal in a Decision-Making Serious Game.” In Proceedings of the 2017 9th International Conference on Virtual Worlds and Games for Serious Applications (VS-Games), 153–156. Red Hook, NY: Curran Associates. Google Scholar logo with link to Google ScholarJercic, Petar, Charlotte Sennersten, and Craig Lindley 2017 “The Effect of Cognitive Load on Physiological Arousal in a Decision-Making Serious Game.” In Proceedings of the 2017 9th International Conference on Virtual Worlds and Games for Serious Applications (VS-Games), 153–156. Red Hook, NY: Curran Associates. Google Scholar logo with link to Google Scholar), and neural activity (Drabant et al. 2011Drabant, Emily M., Janice R. Kuo, Wiveka Ramel, et al. 2011 “Experiential, Autonomic, and Neural Responses during Threat Anticipation Vary as a Function of Threat Intensity and Neuroticism.” NeuroImage 55 (1): 401–410. Google Scholar logo with link to Google ScholarDrabant, Emily M., Janice R. Kuo, Wiveka Ramel, et al. 2011 “Experiential, Autonomic, and Neural Responses during Threat Anticipation Vary as a Function of Threat Intensity and Neuroticism.” NeuroImage 55 (1): 401–410. Google Scholar logo with link to Google Scholar). Both the factorial models and the plot revealed that emotional load has a greater effect in listening than in interpreting. This is to be expected, given that participants could attend to the emotional content of the stimuli to a greater degree without carrying out a cognitively challenging activity, namely, production in CI.

Voice parameters revealed a slightly different pattern from that observed in SC. Although cognitive load did not affect the interpreters’ F 0 , emotional load did: participants produced a lower F 0 when interpreting HELC speech than when interpreting LEHC speech. More importantly, the lowest F 0 was recorded during the interpretation of HELC speech. Conversely, there was no difference in voice intensity between speeches of different emotional content and difficulty. Voice intensity only decreased during the interpreting tasks compared to the baseline task; that is, when producing natural, impromptu speech. In this respect, the results for voice parameters align overall with those of previous studies (e.g., Kumcu 2023Kumcu, Alper 2023 “Emotional Language Processing in Bilingualism: Subjective Affect and Prosodic Markers in Simultaneous Interpreting.” Çankaya University Journal of Humanities and Social Sciences 17 (2): 194–211. Google Scholar logo with link to Google ScholarKumcu, Alper 2023 “Emotional Language Processing in Bilingualism: Subjective Affect and Prosodic Markers in Simultaneous Interpreting.” Çankaya University Journal of Humanities and Social Sciences 17 (2): 194–211. Google Scholar logo with link to Google Scholar), which suggest that F 0 is primarily affected by emotional load in interpreting. Evidence suggests that a higher mean F 0 reflects greater fatigue among interpreters, whereas higher F 0 peaks and a wider F 0 range are associated with greater cognitive load (Shao and Defrancq 2025Shao, Zhangminzi, and Bart Defrancq 2025 “Fundamental Frequency as an Acoustic Mirror of Interpreters’ Cognitive States.” Interpreting 27 (1): 1–27. Google Scholar logo with link to Google ScholarShao, Zhangminzi, and Bart Defrancq 2025 “Fundamental Frequency as an Acoustic Mirror of Interpreters’ Cognitive States.” Interpreting 27 (1): 1–27. Google Scholar logo with link to Google Scholar). Our F 0 results suggest a different mechanism, but methodological discrepancies should be considered when comparing the findings. While it could be argued that voice intensity is modulated by cognitive load, this only applies to the difference between interpreting and producing natural speech in the present study. It could also be argued that voice intensity, as a vocal measure of stress, was not precise enough to distinguish between interpreting tasks, as also reported by Kumcu (2023)Kumcu, Alper 2023 “Emotional Language Processing in Bilingualism: Subjective Affect and Prosodic Markers in Simultaneous Interpreting.” Çankaya University Journal of Humanities and Social Sciences 17 (2): 194–211. Google Scholar logo with link to Google ScholarKumcu, Alper 2023 “Emotional Language Processing in Bilingualism: Subjective Affect and Prosodic Markers in Simultaneous Interpreting.” Çankaya University Journal of Humanities and Social Sciences 17 (2): 194–211. Google Scholar logo with link to Google Scholar. Previous studies investigating the effect of emotionality on F 0 have generally shown that higher emotional load and stress are associated with higher F 0 (Johannes et al. 2007Johannes, Bernd, Peter Wittels, Robert Enne, et al. 2007 “Non-Linear Function Model of Voice Pitch Dependency on Physical and Mental Load.” European Journal of Applied Physiology 101 (3): 267–276. Google Scholar logo with link to Google ScholarJohannes, Bernd, Peter Wittels, Robert Enne, et al. 2007 “Non-Linear Function Model of Voice Pitch Dependency on Physical and Mental Load.” European Journal of Applied Physiology 101 (3): 267–276. Google Scholar logo with link to Google Scholar; Pisanski, Nowak, and Sorokowski 2016Pisanski, Katarzyna, Judyta Nowak, and Piotr Sorokowski 2016 “Individual Differences in Cortisol Stress Response Predict Increases in Voice Pitch during Exam Stress.” Physiology & Behavior 163: 234–238. Google Scholar logo with link to Google ScholarPisanski, Katarzyna, Judyta Nowak, and Piotr Sorokowski 2016 “Individual Differences in Cortisol Stress Response Predict Increases in Voice Pitch during Exam Stress.” Physiology & Behavior 163: 234–238. Google Scholar logo with link to Google Scholar), although this is not a universal finding (see Giddens et al. 2013Giddens, Cheryl L., Kirk W. Barron, Jennifer Byrd-Craven, Keith F. Clark, and A. Scott Winter 2013 “Vocal Indices of Stress: A Review.” Journal of Voice 27 (3): 390.e21–390.e29. Google Scholar logo with link to Google ScholarGiddens, Cheryl L., Kirk W. Barron, Jennifer Byrd-Craven, Keith F. Clark, and A. Scott Winter 2013 “Vocal Indices of Stress: A Review.” Journal of Voice 27 (3): 390.e21–390.e29. Google Scholar logo with link to Google Scholar). That said, supposedly higher emotional load was associated with lower pitch in our study. We hypothesise that this discrepancy is primarily due to the nature of the experimental task: interpreting.

In this respect, the participants produced the lowest F 0 values when interpreting the HELC speech, which warrants further explanation. As the source speaker produced the lowest F 0 during the HELC speech, we refer to ‘source-target synchronisation’ (in other words, a type of ‘emotional mimicry’) to explain our findings, as also shown by the correlations between the source and target speeches. This phenomenon reflects vocal convergence, whereby the interpreter subconsciously aligns their output with the speaker’s acoustic features. As the interpreting task in the current study involved exposure to emotionally charged first-person narratives, the interpreters may have engaged in affective alignment, modulating their voice to reflect the emotional tone of the source speaker. This could be a form of empathic attunement or a strategy for engaging the audience, based on what they learnt in the classroom. In support of this, previous studies have identified comparable cross-speaker vocal alignment in emotionally charged contexts (e.g., Korpal and Jasielska 2019Korpal, Paweł, and Aleksandra Jasielska 2019 “Investigating Interpreters’ Empathy: Are Emotions in Simultaneous Interpreting Contagious?” Target 31 (1): 2–24. Google Scholar logo with link to Google ScholarKorpal, Paweł, and Aleksandra Jasielska 2019 “Investigating Interpreters’ Empathy: Are Emotions in Simultaneous Interpreting Contagious?” Target 31 (1): 2–24. Google Scholar logo with link to Google Scholar; Olza 2024Olza, Inés 2024 “Gestural Alignment in Spoken Simultaneous Interpreting: A Mixed-Methods Approach.” Languages 9 (4): 151. Google Scholar logo with link to Google ScholarOlza, Inés 2024 “Gestural Alignment in Spoken Simultaneous Interpreting: A Mixed-Methods Approach.” Languages 9 (4): 151. Google Scholar logo with link to Google Scholar). It is also important to note that the HELC speech was an offensive and disturbing narrative about an individual with a heroin addiction. ML models revealed that this speech exhibited the highest lexical richness and surprisal, indicating emotional spontaneity and an unpredictable turn of events.

In terms of self-reported measures, participants reported lower positive affect after interpreting any type of speech compared to the baseline task. They also reported higher negative affect after interpreting the HEHC speech than at the baseline task, and higher negative affect than after interpreting all other speech types except the HELC speech. In summary, interpreting both high-emotion speeches increased self-reported negative affect. However, the HEHC speech produced the lowest positive affect and the highest negative affect. Self-reported affect was consistent with findings from voice parameters, in that emotional load modulated negative affect after interpreting, as expected given the nature of the PANAS. As with the results from voice parameters, previous findings were replicated, showing that interpreting emotionally negative sad speeches increased negative affect (Korpal and Jasielska 2019Korpal, Paweł, and Aleksandra Jasielska 2019 “Investigating Interpreters’ Empathy: Are Emotions in Simultaneous Interpreting Contagious?” Target 31 (1): 2–24. Google Scholar logo with link to Google ScholarKorpal, Paweł, and Aleksandra Jasielska 2019 “Investigating Interpreters’ Empathy: Are Emotions in Simultaneous Interpreting Contagious?” Target 31 (1): 2–24. Google Scholar logo with link to Google Scholar; Kumcu 2023Kumcu, Alper 2023 “Emotional Language Processing in Bilingualism: Subjective Affect and Prosodic Markers in Simultaneous Interpreting.” Çankaya University Journal of Humanities and Social Sciences 17 (2): 194–211. Google Scholar logo with link to Google ScholarKumcu, Alper 2023 “Emotional Language Processing in Bilingualism: Subjective Affect and Prosodic Markers in Simultaneous Interpreting.” Çankaya University Journal of Humanities and Social Sciences 17 (2): 194–211. Google Scholar logo with link to Google Scholar), and interpreting as a task overall decreased positive affect compared to producing impromptu speech (Kumcu 2023Kumcu, Alper 2023 “Emotional Language Processing in Bilingualism: Subjective Affect and Prosodic Markers in Simultaneous Interpreting.” Çankaya University Journal of Humanities and Social Sciences 17 (2): 194–211. Google Scholar logo with link to Google ScholarKumcu, Alper 2023 “Emotional Language Processing in Bilingualism: Subjective Affect and Prosodic Markers in Simultaneous Interpreting.” Çankaya University Journal of Humanities and Social Sciences 17 (2): 194–211. Google Scholar logo with link to Google Scholar).

That said, the finding that anxiety (as measured by the STAI) and mental demand (as measured by the NASA-TLX) did not differ significantly across CI tasks, despite earlier reports suggesting otherwise (e.g., Moser-Mercer 2005Moser-Mercer, Barbara 2005 “Remote Interpreting: The Crucial Role of Presence.” Bulletin VALS-ASLA 81: 73–97.Google Scholar logo with link to Google ScholarMoser-Mercer, Barbara 2005 “Remote Interpreting: The Crucial Role of Presence.” Bulletin VALS-ASLA 81: 73–97.Google Scholar logo with link to Google Scholar; Liu, Zheng, and Zhou 2019Liu, Yanmei, Binghan Zheng, and Hao Zhou 2019 “Measuring the Difficulty of Text Translation: The Combination of Text-Focused and Translator-Oriented Approaches.” Target 31 (1): 125–149. Google Scholar logo with link to Google ScholarLiu, Yanmei, Binghan Zheng, and Hao Zhou 2019 “Measuring the Difficulty of Text Translation: The Combination of Text-Focused and Translator-Oriented Approaches.” Target 31 (1): 125–149. Google Scholar logo with link to Google Scholar; Zhao and Lin 2025Zhao, Nan, and Yumeng Lin 2025 “The Roles of Language Proficiency, Working Memory, and Anxiety in Speech Error Repairs in Consecutive Interpreting.” In Mapping Synergies in Cognitive Research on Multilectal Mediated Communication, edited by Laura Babcock, Raphael Sannholm, and Elisabet Tiselius, special issue of Target 37 (2): 184–212. Google Scholar logo with link to Google ScholarZhao, Nan, and Yumeng Lin 2025 “The Roles of Language Proficiency, Working Memory, and Anxiety in Speech Error Repairs in Consecutive Interpreting.” In Mapping Synergies in Cognitive Research on Multilectal Mediated Communication, edited by Laura Babcock, Raphael Sannholm, and Elisabet Tiselius, special issue of Target 37 (2): 184–212. Google Scholar logo with link to Google Scholar), calls for further discussion. One possible explanation concerns the nature and timing of self-reports. The NASA-TLX and STAI were administered after task completion, thereby capturing participants’ overall appraisal of their experience and performance rather than moment-to-moment affective fluctuations during interpreting. In contrast, physiological measures such as GSR reflect core affective arousal in real time and can index the immediate physiological correlates of stress responses. It is therefore plausible that transient increases in arousal occurred in response to cognitively demanding segments but were not sufficiently intense or enduring to be consciously perceived as higher anxiety or mental demand in retrospective self-assessments. In light of the conceptual distinction between affect, stress, and anxiety outlined in Section 1, these findings point to an important pattern. Core affective arousal appears to have fluctuated dynamically with task difficulty. However, participants’ subjective interpretations of that arousal (whether they labelled it as stress or anxiety) remained relatively stable. Future research combining concurrent physiological and self-report measures may better capture how interpreters experience and construe affective and cognitive load during performance.

Lastly, none of the measures in the study predicted CI accuracy, which may seem unexpected at first glance. It is important to reiterate that, unlike accuracy, skin conductance captures real-time physiological responses to momentary stress. Because arousal fluctuates from one moment to the next, whereas accuracy reflects a cumulative end product, these brief physiological changes may not translate into detectable differences in overall performance. Another possible explanation is that participants may have employed different coping or regulatory strategies, so that elevated arousal did not uniformly impair or enhance performance. The absence of perceived difficulty differences between higher and lower cognitive load speeches, despite objective NLP-based evidence of greater linguistic and semantic complexity, suggests that interpreters may have normalised cognitive effort through adaptive strategies, thereby stabilising performance outcomes. Some interpreters may be more skilled at managing stress or may even perform better under moderate levels of arousal. This point is especially relevant given the small sample size, which may have masked individual differences in coping and resilience. Furthermore, students might show greater variability in performance stability under stress than experienced professionals. Since our results suggest that cognitive and emotional load fluctuate dynamically during interpreting, it is plausible that participants adapted their strategies online to maintain stable performance despite transient increases in physiological arousal. Future research could extend this work by testing professional interpreters or by incorporating time-resolved performance measures that track accuracy as it unfolds. Studies could also include individual-difference variables, such as working memory capacity or emotion-regulation ability, to better explain why interpreters differ in how they adapt to fluctuating demands.

In terms of text profiling using NLP/ML methods, future studies could use time windows within speeches to capture dynamic changes in metrics such as windowed surprisal, which tracks shifts in predictability across short segments. Furthermore, GAMMs or hierarchical mixed models could incorporate fine-grained predictors such as the surprisal or semantic dispersion of the current segment, aligned with corresponding physiological or acoustic-prosodic data, allowing the analysis to move beyond aggregated averages toward more detailed, time-resolved modelling. Future studies could also consider ecological add-ons. For example, they could combine text profiling with delivery features (e.g., speech rate and disfluencies) and acoustic measures. In any case, we recommend that future studies utilising NLP/ML techniques employ tailor-made, monolingual embeddings rather than general, multilingual ones for source-language profiling and adhere to best practices for transparency and reproducibility. This should include reporting model and version details, sharing code and stimulus-level metric tables, and publishing speech audio and transcripts with derived features, to facilitate cumulative work.

Similarly, a more detailed analysis of voice parameters within smaller speech segments during the interpreting process could provide a more comprehensive picture of the changing nature of interpreting. Measuring skin responses during passive listening as a control condition could strengthen the argument for a difference between passive listening and listening for interpreting. The emotional stimuli in the study were exclusively negative in valence. Thus, the findings may not generalise to positive emotional content; however, future studies can use positively valenced stimuli to test for valence-specific effects. Lastly, it is important to note that the present study primarily measured interpreters’ effort rather than load per se. Load was manipulated through task properties, while the dependent measures reflected interpreters’ responses to these demands.

5.Conclusion

The present study challenges the assumption that emotional content is the primary source of stress in interpreting, demonstrating that cognitive load, rather than emotional content, is the primary trigger of physiological arousal. Emotional load clearly played a role, albeit to a lesser extent, but it manifested differently from cognitive load and left a distinct empirical signature. In this respect, we propose a three-system model of interpreter response, comprising arousal, measured via skin conductance as a marker of cognitive strain; voice parameters, as a marker of emotional modulation; and self-reports, as a marker of subjective affect. This should not be interpreted to mean that GSR is exclusive to cognitive load. On the contrary, GSR’s relatively heightened sensitivity to cognitive load enabled us to differentiate the relative influence of cognitive and emotional demands when tested within the same experiment. We also emphasise the importance of temporal dynamics (Rabinovich et al. 2012Rabinovich, Mikhail I., Valentin S. Afraimovich, Christian Bick, and Pablo Varona 2012 “Information Flow Dynamics in the Brain.” Physics of Life Reviews 9 (1): 51–73. Google Scholar logo with link to Google ScholarRabinovich, Mikhail I., Valentin S. Afraimovich, Christian Bick, and Pablo Varona 2012 “Information Flow Dynamics in the Brain.” Physics of Life Reviews 9 (1): 51–73. Google Scholar logo with link to Google Scholar) and ecological cognition (Spivey and Dale 2006Spivey, Michael J., and Rick Dale 2006 “Continuous Dynamics in Real-Time Cognition.” Current Directions in Psychological Science 15 (5): 207–211. Google Scholar logo with link to Google ScholarSpivey, Michael J., and Rick Dale 2006 “Continuous Dynamics in Real-Time Cognition.” Current Directions in Psychological Science 15 (5): 207–211. Google Scholar logo with link to Google Scholar) in interpreting, noting that stress builds incrementally over time. This supports the need for dynamic models rather than static cognitive/emotional processing models in interpreting. Our findings also have implications for interpreting research and training. By demonstrating that cognitive load is a stronger driver of physiological stress than emotional load, the study highlights the continuous need to prioritise cognitive management strategies in interpreter education. However, emotional regulation should not be overlooked, particularly in high-stakes settings such as mental healthcare interpreting.

Funding

Open Access publication of this article was funded through a Transformative Agreement with Bilkent University.

References

Adler, Aleksandra
2023 Cognitive Load in Dialogue Interpreting: Experience and Directionality. PhD diss. Stockholm University. http://​urn​.kb​.se​/resolve​?urn​=urn:nbn:se:su:diva​-220765
①
2024 “Effects of Experience and Directionality on Cognitive Load in Dialogue Interpreting.” Translation, Cognition & Behavior 7 (2): 187–208. Google Scholar logo with link to Google Scholar
①
Amato, Amalia, and Gabriele Mack
2011 “Interpreting the Oscar Night on Italian TV: An Interpreters’ Nightmare?” The Interpreters’ Newsletter 16: 37–60.Google Scholar logo with link to Google Scholar
①
Ateşman, Ender
1997 “Türkçede okunabilirliğin ölçülmesi [Measuring readability in Turkish].” Dil Dergisi 58: 71–74.Google Scholar logo with link to Google Scholar
①
Augustine, Adam A., and Scott H. Hemenover
2009 “On the Relative Effectiveness of Affect Regulation Strategies: A Meta-Analysis.” Cognition and Emotion 23 (6): 1181–1220. Google Scholar logo with link to Google Scholar
①
Baayen, R. Harald, Douglas J. Davidson, and Douglas M. Bates
2008 “Mixed-Effects Modeling with Crossed Random Effects for Subjects and Items.” Journal of Memory and Language 59 (4): 390–412. Google Scholar logo with link to Google Scholar
①
Bachorowski, Jo-Anne, and Michael J. Owren
1995 “Vocal Expression of Emotion: Acoustic Properties of Speech Are Associated with Emotional Intensity and Context.” Psychological Science 6 (4): 219–224. Google Scholar logo with link to Google Scholar
①
Bates, Douglas M., Martin Mächler, Ben Bolker, and Steve Walker
2015 “Fitting Linear Mixed-Effects Models Using lme4.” Journal of Statistical Software 67 (1): 1–48. Google Scholar logo with link to Google Scholar
①
Benedek, Mathias, and Christian Kaernbach
2010 “A Continuous Measure of Phasic Electrodermal Activity.” Journal of Neuroscience Methods 190 (1): 80–91. Google Scholar logo with link to Google Scholar
①
Boersma, Paul, and David Weenink
2022 Praat: Doing Phonetics by Computer. Version 6.3.03. Amsterdam. http://​www​.praat​.org/
①
Boucsein, Wolfram, Don C. Fowles, Sverre Grimnes, Gershon Ben-Shakhar, Walton T. Roth, Michael E. Dawson, and Diane L. Filion
2012 “Publication Recommendations for Electrodermal Measurements.” Psychophysiology 49 (8): 1017–1034. Google Scholar logo with link to Google Scholar
①
Caldwell-Harris, Catherine L., Jimmy Tong, Winvy Lung, and Sinlan Poo
2011 “Physiological Reactivity to Emotional Phrases in Mandarin–English Bilinguals.” International Journal of Bilingualism 15 (3): 329–352. Google Scholar logo with link to Google Scholar
①
Çetinkaya, Gökhan, and Leyla Uzun
2018 “Türkçe ders kitaplarındaki metinlerin okunabilirlik özellikleri [Readability features of texts in Turkish textbooks].” In Türkçe ders kitabı çözümlemeleri [Analyses of Turkish textbooks], edited by Hakan Ülper, 141–156. Ankara: Pegem Akademi.Google Scholar logo with link to Google Scholar
①
Chen, Sijia
2017 “The Construct of Cognitive Load in Interpreting and Its Measurement.” Perspectives 25 (4): 640–657. Google Scholar logo with link to Google Scholar
①
Clark, Herbert H.
1973 “The Language-as-Fixed-Effect Fallacy: A Critique of Language Statistics in Psychological Research.” Journal of Verbal Learning and Verbal Behavior 12 (4): 335–359. Google Scholar logo with link to Google Scholar
①
Crezee, Ineke, Shirley Jülich, and Maria Hayward
2013 “Issues for Interpreters and Professionals Working in Refugee Settings.” Journal of Applied Linguistics and Professional Practice 8 (3): 253–273. Google Scholar logo with link to Google Scholar
①
Daly, Brad, and Cathy J. Chovaz
2020 “Secondary Traumatic Stress: Effects on the Professional Quality of Life of Sign Language Interpreters.” American Annals of the Deaf 165 (3): 353–367. Google Scholar logo with link to Google Scholar
①
Delisle, Alain, Christian Larivière, Daniel Imbeau, and Marie-José Durand
2005 “Physical Exposure of Sign Language Interpreters: Baseline Measures and Reliability Analysis.” European Journal of Applied Physiology 94 (4): 448–460. Google Scholar logo with link to Google Scholar
①
Delizée, Anne, Nathalie Bennoun, Felicia Dutray, Lorine Pierard, and Orest Weber
2024 “Preventing Vicarious Traumatisation in Mental Health Settings: A Training Course for Interpreters.” FITISPos International Journal 11 (1): 28–43. Google Scholar logo with link to Google Scholar
①
Dong, Yanping, Yuhua Liu, and Rendong Cai
2018 “How Does Consecutive Interpreting Training Influence Working Memory: A Longitudinal Study of Potential Links between the Two.” Frontiers in Psychology 9: 875. Google Scholar logo with link to Google Scholar
①
Drabant, Emily M., Janice R. Kuo, Wiveka Ramel, et al.
2011 “Experiential, Autonomic, and Neural Responses during Threat Anticipation Vary as a Function of Threat Intensity and Neuroticism.” NeuroImage 55 (1): 401–410. Google Scholar logo with link to Google Scholar
①
Geiling, Angelika, Christine Knaevelsrud, Maria Böttche, and Nadine Stammel
2021 “Mental Health and Work Experiences of Interpreters in the Mental Health Care of Refugees: A Systematic Review.” Frontiers in Psychiatry 12: 710789. Google Scholar logo with link to Google Scholar
①
2022 “Psychological Distress, Exhaustion, and Work-Related Correlates among Interpreters Working in Refugee Care: Results of a Nationwide Online Survey in Germany.” European Journal of Psychotraumatology 13 (1): 2046954. Google Scholar logo with link to Google Scholar
①
Giddens, Cheryl L., Kirk W. Barron, Jennifer Byrd-Craven, Keith F. Clark, and A. Scott Winter
2013 “Vocal Indices of Stress: A Review.” Journal of Voice 27 (3): 390.e21–390.e29. Google Scholar logo with link to Google Scholar
①
Gieshoff, Anne Catherine, and Michaela Albl-Mikasa
2024 “Interpreting Accuracy Revisited: A Refined Approach to Interpreting Performance Analysis.” Perspectives 32 (2): 210–228. Google Scholar logo with link to Google Scholar
①
Gieshoff, Anne Catherine, and Andrea Hunziker Heeb
2023 “Cognitive Load and Cognitive Effort: Probing the Psychological Reality of a Conceptual Difference.” Translation, Cognition & Behavior 6 (1): 3–28. Google Scholar logo with link to Google Scholar
①
Gieshoff, Anne Catherine, Caroline Lehr, and Andrea Hunziker Heeb
2021 “Stress, Cognitive, Emotional and Ergonomic Demands in Interpreting and Translation: A Review of Physiological Studies.” Cognitive Linguistic Studies 8 (2): 404–439. Google Scholar logo with link to Google Scholar
①
Gumul, Ewa
2021 “Reporting Stress in Simultaneous Interpreting: The Analysis of Trainee Interpreters’ Retrospective Reports and Outputs.” Onomázein NE8: 16–42.Google Scholar logo with link to Google Scholar
① ②
Han, Chao
2018 “Using Rating Scales to Assess Interpretation: Practices, Problems and Prospects.” Interpreting 20 (1): 63–101. Google Scholar logo with link to Google Scholar
①
Hancox, Jennifer A., Clare F. McKiernan, Alice L. Martin, Jon Tomas, and John I. MacArtney
2023 “The Emotional Effects on Professional Interpreters of Interpreting Palliative Care Conversations for Adult Patients: A Rapid Review.” Palliative Medicine 37 (7): 931–946. Google Scholar logo with link to Google Scholar
①
Harris, Catherine L., Ayşe Ayçiçeği, and Jean Berko Gleason
2003 “Taboo Words and Reprimands Elicit Greater Autonomic Reactivity in a First Language than in a Second Language.” Applied Psycholinguistics 24 (4): 561–579. Google Scholar logo with link to Google Scholar
①
Hart, Sandra G., and Lowell E. Staveland
1988 “Development of NASA-TLX (Task Load Index): Results of Empirical and Theoretical Research.” In Human Mental Workload, edited by Peter A. Hancock and Najmedin Meshkati, 139–183. Amsterdam: Elsevier.Google Scholar logo with link to Google Scholar
①
Hsieh, Elaine
2006 “Conflicts in How Interpreters Manage Their Roles in Provider–Patient Interactions.” Social Science & Medicine 62 (3): 721–730. Google Scholar logo with link to Google Scholar
①
Jercic, Petar, Charlotte Sennersten, and Craig Lindley
2017 “The Effect of Cognitive Load on Physiological Arousal in a Decision-Making Serious Game.” In Proceedings of the 2017 9th International Conference on Virtual Worlds and Games for Serious Applications (VS-Games), 153–156. Red Hook, NY: Curran Associates. Google Scholar logo with link to Google Scholar
①
Johannes, Bernd, Peter Wittels, Robert Enne, et al.
2007 “Non-Linear Function Model of Voice Pitch Dependency on Physical and Mental Load.” European Journal of Applied Physiology 101 (3): 267–276. Google Scholar logo with link to Google Scholar
①
Kapucu, Aycan, Aslı Kılıç, Yıldız Özkılıç, and Bengisu Sarıbaz
2021 “Turkish Emotional Word Norms for Arousal, Valence, and Discrete Emotion Categories.” Psychological Reports 124 (1): 188–209. Google Scholar logo with link to Google Scholar
①
Kesgin, H. Toprak, M. Kaan Yuce, Eren Dogan, M. Egemen Uzun, Atahan Uz, H. Emre Seyrek, Ahmed Zeer, and M. Fatih Amasyali
2024 “Introducing cosmosGPT: Monolingual Training for Turkish Language Models.” Preprint, arXiv. https://​arxiv​.org​/abs​/2404​.17336
①
Kindermann, David, Carolin Schmid, Cassandra Derreza-Greeven, et al.
2017 “Prevalence of and Risk Factors for Secondary Traumatization in Interpreters for Refugees: A Cross-Sectional Study.” Psychopathology 50 (4): 262–272. Google Scholar logo with link to Google Scholar
①
Klonowicz, Tatiana
1990 “A Psychophysiological Assessment of Simultaneous Interpreting: The Interaction of Individual Differences and Mental Workload.” Polish Psychological Bulletin 21: 37–48.Google Scholar logo with link to Google Scholar
①
1994 “Putting One’s Hart into Simultaneous Interpretation.” In Bridging the Gap: Empirical Research in Simultaneous Interpretation, edited by Sylvie Lambert and Barbara Moser-Mercer, 213–224. Amsterdam: John Benjamins. Google Scholar logo with link to Google Scholar
①
Korpal, Paweł
2016 “Interpreting as a Stressful Activity: Physiological Measures of Stress in Simultaneous Interpreting.” Poznan Studies in Contemporary Linguistics 52 (2): 297–316. Google Scholar logo with link to Google Scholar
① ②
2021 “Stress and Emotion in Conference Interpreting.” In The Routledge Handbook of Conference Interpreting, edited by Michaela Albl-Mikasa and Elisabet Tiselius, 401–413. London: Routledge. Google Scholar logo with link to Google Scholar
①
Korpal, Paweł, and Aleksandra Jasielska
2019 “Investigating Interpreters’ Empathy: Are Emotions in Simultaneous Interpreting Contagious?” Target 31 (1): 2–24. Google Scholar logo with link to Google Scholar
① ② ③ ④ ⑤ ⑥
Korpal, Paweł, Katarzyna Jankowiak, and Łukasz D. Kaczmarek
2026 “Emotional Content Affects Interpreters: Insights from Psychophysiological, Behavioral, and Self-Report Measures.” Interpreting 28 (1): 1–29. Google Scholar logo with link to Google Scholar
①
Kuang, Huolingxiao, and Binghan Zheng
2023 “Note-Taking Effort in Video Remote Interpreting: Effects of Source Speech Difficulty and Interpreter Work Experience.” Perspectives 31 (4): 724–744. Google Scholar logo with link to Google Scholar
①
Kumcu, Alper
2023 “Emotional Language Processing in Bilingualism: Subjective Affect and Prosodic Markers in Simultaneous Interpreting.” Çankaya University Journal of Humanities and Social Sciences 17 (2): 194–211. Google Scholar logo with link to Google Scholar
① ② ③ ④ ⑤ ⑥ ⑦
Kumcu, Alper, and Asiye Öztürk
2023 “Visual Mental Imagery and Verbal Working Memory: Evidence from Consecutive Interpreting.” Journal of Cognitive Psychology 35 (5): 545–560. Google Scholar logo with link to Google Scholar
①
Kurz, Ingrid
2002 “Physiological Stress Responses during Media and Conference Interpreting.” In Interpreting in the 21st Century: Challenges and Opportunities, edited by Giuliana Garzone and Maurizio Viezzi, 195–202. Amsterdam: John Benjamins. Google Scholar logo with link to Google Scholar
① ② ③
2003 “Physiological Stress during Simultaneous Interpreting: A Comparison of Experts and Novices.” The Interpreters’ Newsletter 12: 51–67.Google Scholar logo with link to Google Scholar
① ②
Kuznetsova, Alexandra, Per B. Brockhoff, and Rune Haubo Bojesen Christensen
2017 “lmerTest Package: Tests in Linear Mixed Effects Models.” Journal of Statistical Software 82 (13): 1–26. Google Scholar logo with link to Google Scholar
①
Lai, Miranda, and Georgina Heydon
2015 “Vicarious Trauma among Interpreters.” International Journal of Interpreter Education 7 (1): 3–23.Google Scholar logo with link to Google Scholar
①
Leanza, Yvan, Sophie Pointurier, and Laurie Duchesne
2025 “Interpreting Is an Emotional Effort: Public Service Interpreters’ Coping Strategies in Emotionally Charged Situations.” Interpreting and Society 5 (1): 3–26. Google Scholar logo with link to Google Scholar
①
Lee, Jieun
2008 “Rating Scales for Interpreting Performance Assessment.” The Interpreter and Translator Trainer 2 (2): 165–184. Google Scholar logo with link to Google Scholar
①
Li, Saihong, Yifang Wang, and Yubo Zhou Rasmussen
2022 “Studying Interpreters’ Stress in Crisis Communication: Evidence from Multimodal Technology of Eye-Tracking, Heart Rate and Galvanic Skin Response.” The Translator 28 (4): 468–488. Google Scholar logo with link to Google Scholar
① ② ③
Liu, Yanmei, Binghan Zheng, and Hao Zhou
2019 “Measuring the Difficulty of Text Translation: The Combination of Text-Focused and Translator-Oriented Approaches.” Target 31 (1): 125–149. Google Scholar logo with link to Google Scholar
①
Mackintosh, Jennifer
2003 “The AIIC Workload Study.” Forum 1 (2): 189–214. Google Scholar logo with link to Google Scholar
①
Mellinger, Christopher D.
2022 “Cognitive Behaviour during Consecutive Interpreting: Describing the Notetaking Process.” Translation & Interpreting 14 (2): 103–119. Google Scholar logo with link to Google Scholar
①
Moser-Mercer, Barbara, Alexander Künzli, and Marina Korac
1998 “Prolonged Turns in Interpreting: Effects on Quality, Physiological and Psychological Stress (Pilot Study).” Interpreting 3 (1): 47–64. Google Scholar logo with link to Google Scholar
①
Moser-Mercer, Barbara
2005 “Remote Interpreting: The Crucial Role of Presence.” Bulletin VALS-ASLA 81: 73–97.Google Scholar logo with link to Google Scholar
① ②
Muñoz Martín, Ricardo, and Ana María Rojo López
2025 “Introduction: Research Methods in Cognitive Translation and Interpreting Studies.” In Research Methods in Cognitive Translation and Interpreting Studies, edited by Ana María Rojo López and Ricardo Muñoz Martín, 1–20. Amsterdam: John Benjamins. Google Scholar logo with link to Google Scholar
①
Olalla-Soler, Christian, Nicoletta Spinolo, and Ricardo Muñoz Martín
2023 “Under Pressure? A Study of Heart Rate and Heart-Rate Variability Using SmarTerp.” HERMES: Journal of Language and Communication in Business 63: 119–142.Google Scholar logo with link to Google Scholar
①
Olza, Inés
2024 “Gestural Alignment in Spoken Simultaneous Interpreting: A Mixed-Methods Approach.” Languages 9 (4): 151. Google Scholar logo with link to Google Scholar
①
Park, Elyse R., Jan E. Mutchler, Giselle Perez, et al.
2017 “Coping and Resiliency Enhancement Program (CARE): A Pilot Study for Interpreters in Cancer Care.” Psycho-Oncology 26 (8): 1181–1190. Google Scholar logo with link to Google Scholar
①
Pashler, Harold
1994 “Dual-Task Interference in Simple Tasks: Data and Theory.” Psychological Bulletin 116 (2): 220–244. Google Scholar logo with link to Google Scholar
①
Pisanski, Katarzyna, Agata Groyecka-Bernard, and Piotr Sorokowski
2021 “Human Voice Pitch Measures Are Robust across a Variety of Speech Recordings: Methodological and Theoretical Implications.” Biology Letters 17 (9): 20210356. Google Scholar logo with link to Google Scholar
①
Pisanski, Katarzyna, Judyta Nowak, and Piotr Sorokowski
2016 “Individual Differences in Cortisol Stress Response Predict Increases in Voice Pitch during Exam Stress.” Physiology & Behavior 163: 234–238. Google Scholar logo with link to Google Scholar
①
Puerini, Sara, and Ricardo Muñoz Martín
2025 “Beyond Readability Formulas: Text Profiling in Cognitive Translation & Interpreting Studies.” Paper presented at Fifth International Conference on Translation, Interpreting and Cognition, Kristiansand, Norway. June 2–4.
①
R Core Team
2025 R: A Language and Environment for Statistical Computing. Vienna: R Foundation for Statistical Computing. https://​www​.r​-project​.org/
①
Rabinovich, Mikhail I., Valentin S. Afraimovich, Christian Bick, and Pablo Varona
2012 “Information Flow Dynamics in the Brain.” Physics of Life Reviews 9 (1): 51–73. Google Scholar logo with link to Google Scholar
①
Reimers, Nils, and Iryna Gurevych
2019 “Sentence-BERT: Sentence Embeddings Using Siamese BERT-Networks.” In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), edited by Kentaro Inui, Jing Jiang, Vincent Ng, and Xiaojun Wan, 3982–3992. Stroudsburg, PA: Association for Computational Linguistics. Google Scholar logo with link to Google Scholar
①
Riccardi, Alessandra, Guido Marinuzzi, and Stefano Zecchin
1998 “Interpretation and Stress.” The Interpreters’ Newsletter 8: 93–106.Google Scholar logo with link to Google Scholar
①
Richardson, Katherine M., and Hannah R. Rothstein
2008 “Effects of Occupational Stress Management Intervention Programs: A Meta-Analysis.” Journal of Occupational Health Psychology 13 (1): 69–93. Google Scholar logo with link to Google Scholar
①
Rojo López, Ana María, and Ana Isabel Foulquié Rubio
2024 “Interpreting, Affect, and Emotion.” In The Routledge Handbook of Interpreting and Cognition, edited by Christopher D. Mellinger, 307–323. London: Routledge. Google Scholar logo with link to Google Scholar
① ②
Rojo López, Ana María, Ana Isabel Foulquié Rubio, Laura Espín López, and Francisco Martínez Sánchez
2021 “Analysis of Speech Rhythm and Heart Rate as Indicators of Stress on Student Interpreters.” Perspectives 29 (4): 591–607. Google Scholar logo with link to Google Scholar
① ②
Rojo López, Ana María, Marina Ramos Caro, Daria Patalas, and Aleksandra Jasielska
2025 “Psychophysiological Methods.” In Research Methods in Cognitive Translation and Interpreting Studies, edited by Ana María Rojo López and Ricardo Muñoz Martín, 256–278. Amsterdam: John Benjamins. Google Scholar logo with link to Google Scholar
①
Roziner, Ilan, and Miriam Shlesinger
2010 “Much Ado about Something Remote: Stress and Performance in Remote Interpreting.” Interpreting 12 (2): 214–247. Google Scholar logo with link to Google Scholar
①
Scherer, Klaus R.
1986 “Voice, Stress, and Emotion.” In Dynamics of Stress, edited by Mortimer H. Appley and Richard Trumbull, 157–179. New York: Plenum Press. Google Scholar logo with link to Google Scholar
①
Schmaußer, Maximilian, and Sylvain Laborde
2023 “Tonic and Phasic Cardiac Vagal Activity Predict Cognitive-Affective Processing in an Emotional Stop-Signal Task.” International Journal of Psychophysiology 191: 9–18. Google Scholar logo with link to Google Scholar
①
Seeber, Kilian G.
2011 “Cognitive Load in Simultaneous Interpreting: Existing Theories — New Models.” Interpreting 13 (2): 176–204. Google Scholar logo with link to Google Scholar
①
Shao, Zhangminzi, and Bart Defrancq
2025 “Fundamental Frequency as an Acoustic Mirror of Interpreters’ Cognitive States.” Interpreting 27 (1): 1–27. Google Scholar logo with link to Google Scholar
①
Simms, James V., Alexandra R. Thelan, Sarah E. Domoff, and Elizabeth A. Meadows
2021 “An Examination of Vicarious Trauma among Refugee Mental Health Interpreters.” Occupational Health Science 5 (4): 581–601. Google Scholar logo with link to Google Scholar
①
Sondhi, Savita, Munna Khan, Ritu Vijay, Ashok K. Salhan, and Satish Chouhan
2015 “Acoustic Analysis of Speech under Stress.” International Journal of Bioinformatics Research and Applications 11 (5): 417–432. Google Scholar logo with link to Google Scholar
①
Spielberger, Charles D., Richard L. Gorsuch, Robert E. Lushene, Paul R. Vagg, and Gerard A. Jacobs
1983 Manual for the State-Trait Anxiety Inventory (Form Y1–Y2). Palo Alto: Consulting Psychologists Press.Google Scholar logo with link to Google Scholar
① ②
Spivey, Michael J., and Rick Dale
2006 “Continuous Dynamics in Real-Time Cognition.” Current Directions in Psychological Science 15 (5): 207–211. Google Scholar logo with link to Google Scholar
①
Sun, Sanjun, and Gregory M. Shreve
2014 “Measuring Translation Difficulty: An Empirical Study.” Target 26 (1): 98–127. Google Scholar logo with link to Google Scholar
①
Tiselius, Elisabet, and Birgitta Englund Dimitrova
2023 “Testing the Working Memory Capacity of Dialogue Interpreters.” Across Languages and Cultures 24 (2): 163–180. Google Scholar logo with link to Google Scholar
①
Walczyński, Marcin
2021 “Polish–English Certified Interpreters in Psycho-Affectively Challenging Contexts.” Current Trends in Translation Teaching and Learning E 8: 394–448. Google Scholar logo with link to Google Scholar
①
Watson, David, Lee Anna Clark, and Auke Tellegen
1988 “Development and Validation of Brief Measures of Positive and Negative Affect: The PANAS Scales.” Journal of Personality and Social Psychology 54 (6): 1063–1070. Google Scholar logo with link to Google Scholar
①
Winter, Bodo
2013 “Linear Models and Linear Mixed Effects Models in R with Linguistic Applications.” Preprint, arXiv. http://​arxiv​.org​/abs​/1308​.5499
①
Wood, Simon N.
2011 “Fast Stable Restricted Maximum Likelihood and Marginal Likelihood Estimation of Semiparametric Generalized Linear Models.” Journal of the Royal Statistical Society (B) 73 (1): 3–36. Google Scholar logo with link to Google Scholar
①
Yan, Hao, Yi Zhang, Yanqin Feng, et al.
2024 “Assessing Mental Demand in Consecutive Interpreting: Insights from an fNIRS Study.” Acta Psychologica 243: 104132. Google Scholar logo with link to Google Scholar
① ②
Yang, Po-Chou, I-Mei Lin, and Da-Wei Wu
2025 “Effects of Heart Rate Variability Biofeedback on Enhancing Self-Efficacy, Quality of Life and Six-Minute Walking Test in Patients with Chronic Obstructive Pulmonary Disease.” Applied Psychophysiology and Biofeedback 50: 403–416. Google Scholar logo with link to Google Scholar
①
Zhao, Nan, and Yumeng Lin
2025 “The Roles of Language Proficiency, Working Memory, and Anxiety in Speech Error Repairs in Consecutive Interpreting.” In Mapping Synergies in Cognitive Research on Multilectal Mediated Communication, edited by Laura Babcock, Raphael Sannholm, and Elisabet Tiselius, special issue of Target 37 (2): 184–212. Google Scholar logo with link to Google Scholar
①

Address for correspondence

Duygu Duman

Department of Translation and Interpreting

İhsan Doğramacı Bilkent University

Main Campus, LC Building

Üniversiteler Neighbourhood, Bilkent 8th Avenue

06800 ANKARA

Türkiye

[email protected]
ORCiD logo with linkhttps://orcid.org/0000-0003-3120-0577

Co-author information

Alper Kumcu
Hacettepe University
[email protected]
ORCiD logo with linkhttps://orcid.org/0000-0003-0844-3562
 
Authentication
Mobile Menu Logo with link to supplementary files background Layer 1 jbe-platform.quadraat-sans-bold jbe-platform.quadraat-sans-bold-dark jbe-platform.quadraat-sans-bold-light jbe-platform.quadraat-sans-bold prag Twitter_Logo_Blue