Introduction
Having been extensively discussed in the past decade, the issue of interpreter advantage is one of the most representative examples illustrating the plasticity of human neural and cognitive systems. Interpreter advantages are usually manifested as superior performance for interpreters than matched general bilinguals in tasks gauging cognitive control abilities. The rationale behind the advantage is that the highly complex and cognitively demanding nature of interpreting is assumed to impose a heavy burden on interpreters, which may contribute to the enhancement of their cognitive control abilities or skills. Many studies have been devoted to the investigation of interpreter advantages, mainly covering the three basic executive functions specified in Diamond (Reference Diamond2013), i.e., working memory, inhibitory control, and cognitive flexibility (see Dong & Zhong, Reference Dong, Zhong, Schwieter and Paradis2019; García et al., Reference García, Muñoz and Kogan2020 for reviews). These studies have found interpreter advantages in cognitive flexibility (e.g., Yudes et al., Reference Yudes, Macizo and Bajo2011; Zhao & Dong, Reference Zhao and Dong2020), and working memory (see Ghiselli, Reference Ghiselli2022; Wen & Dong, Reference Wen and Dong2019 for meta-analyses). Although scant evidence was obtained for an advantage in inhibitory control with behavioral measures (e.g., Dong & Liu, Reference Dong and Liu2016; Morales et al., Reference Morales, Padilla, Gómez-Ariza and Bajo2015; but see Henrard & Van Daele, Reference Henrard and Van Daele2017), an Event-related Potentials (ERP) study (Dong & Zhong, Reference Dong and Zhong2017) did offer robust neurological evidence.
Despite many studies devoted to elucidating the interpreter advantage issue, at least two limitations hindered us from depicting a more comprehensive picture of the issue. The first limitation is insufficient attention paid to the coordination skill, which is quite essential to the completion of an interpreting task consisting of multiple simultaneously ongoing processes. Research on interpreter coordination advantage may shed light upon the multitasking nature of interpreting and may help better explore the consequences of interpreting experience (see Dong, Reference Dong, Derreira and Schwieter2023 for a recent comprehensive review). Another limitation is a lack of research on the neural correlates of the interpreter advantage, which is helpful in investigating the underlying mechanisms of the advantage. Neurological techniques like event-related potentials may help unveil hidden information behind mere behavioral responses and expound the issue along the time course of task processing.
Interpreter advantage in coordination skill
As discussed above, studies on the interpreter advantage in the three basic executive functions cannot fully account for the major characteristics of interpreting. One of the underexplored characteristics reflecting the intense and unique nature of an interpreting task is multitasking. Interpreters have to deal with multiple processes or subtasks proceeding within a very limited time period, during which the coordination skill is indispensable (Dong & Li, Reference Dong and Li2020; Gile, 1995/Reference Gile2009; Strobach et al., Reference Strobach, Salminen, Karbach and Schubert2014). As shown in Gile’s (1995/Reference Gile2009) effort model, coordination plays an important role in both simultaneous interpreting (SI) and consecutive interpreting (CI), i.e., interpreters have to coordinate various subtasks or processes such as listening and speaking in SI, listening and note-taking, or speaking and note-reading in CI. Therefore, the skill of coordinating multiple tasks or processes may be frequently exercised in interpreting training or professional interpreting experience, ultimately eliciting an interpreter advantage in coordination.
Nevertheless, the issue of interpreter advantage in coordination remains underexplored, especially the advantage in early stages of interpreting training. Up till now, only one study (Zhong & Dong, Reference Dong, Derreira and Schwieter2023) tapped into these stages. Zhong and Dong (Reference Dong, Derreira and Schwieter2023) asked CI students and general bilingual controls to perform a dual-task. In the single-task condition, participants determined the pitch of a tone (auditory task) or the size of a triangle (visual task), while in the dual-task condition, the auditory task (Task 1) was first presented, and then the visual task (Task 2), with different stimulus onset asynchrony (SOA) in between. The results showed that the interpreting students at the intermediate stage (Chinese-English graduate students with around two years’ training), but not those at the beginning stage (half a year), exhibited smaller dual-task costs in Task 2 but not in Task 1. These results suggest an interpreter advantage in coordination which may only appear at a later stage of interpreting training.
Apart from Zhong and Dong (Reference Dong, Derreira and Schwieter2023), the remaining four studies on the issue of interpreters’ coordination advantage (Becker et al., Reference Becker, Schubert, Strobach, Gallinat and Kühn2016; Morales et al., Reference Morales, Padilla, Gómez-Ariza and Bajo2015; Padilla et al., Reference Padilla, Bajo and Macizo2005; Strobach et al., Reference Strobach, Becker, Schubert and Kühn2015) all recruited professional interpreters, with two of the studies finding positive evidence and two of them finding none. Strobach et al. (Reference Strobach, Becker, Schubert and Kühn2015) found better performance in Task 1 and Task 2 for simultaneous interpreters (SIs) who had equivalent CI and translation experience to their non-SI controls, also suggesting an interpreter advantage in coordination. Becker et al. (Reference Becker, Schubert, Strobach, Gallinat and Kühn2016), with the fMRI technique measuring the same participant sample, found that the advantage in Task 2 may be associated with SIs’ better-developed frontal pole cluster, in which SIs exhibited better global efficiency, higher node degree (the number of edges to which the node is connected), and stronger connectivity with left middle temporal gyrus than the controls. Morales et al. (Reference Morales, Padilla, Gómez-Ariza and Bajo2015) and Padilla et al. (Reference Padilla, Bajo and Macizo2005), however, did not find any evidence for an interpreter advantage in coordination. In Morales et al. (Reference Morales, Padilla, Gómez-Ariza and Bajo2015), SIs did not differ from general bilingual controls in the performance of a dual-task consisting of an auditory N-back task and a visual N-back task. In Padilla et al. (Reference Padilla, Bajo and Macizo2005), there were no significant differences among professional interpreters, professional noninterpreters, and psychology students of high working memory span in the performance of a dual-task composed of a visual tracking task and a free recall task.
The above review shows that experimental tasks and indices for coordination are probably critical in investigating the interpreter’s advantage in coordination. Studies showing no evidence for the advantage employed tasks not requiring speeded responses (Padilla et al., Reference Padilla, Bajo and Macizo2005) or an index showing relatively less information on speeded responses, i.e., accuracy (Morales et al., Reference Morales, Padilla, Gómez-Ariza and Bajo2015), which, according to Strobach et al. (Reference Strobach, Becker, Schubert and Kühn2015), are probably not sensitive enough for observing the coordination advantage. The psychological refractory period (PRP) dual-task (employed in Becker et al., Reference Becker, Schubert, Strobach, Gallinat and Kühn2016, Strobach et al., Reference Strobach, Becker, Schubert and Kühn2015 and Zhong & Dong, Reference Dong, Derreira and Schwieter2023), with speeded responses and response time (RT) directly showing the efficiency of speeded responses, is probably more appropriate for this purpose.
The PRP dual-task and its neurological measures
PRP refers to the psychological refractory period, a term describing the dual-task situation in which the decrease of SOA between Task 1 (the task appeared first) and Task 2 results in longer delay of Task 2 RT (Pashler, Reference Pashler1994). And thus, many PRP dual-tasks (e.g., Becker et al., Reference Becker, Schubert, Strobach, Gallinat and Kühn2016; Mittelstädt et al., Reference Mittelstädt, Mackenzie, Leuthold and Miller2022; Osman & Moore, Reference Osman and Moore1993; Strobach et al., Reference Strobach, Becker, Schubert and Kühn2015; Ulrich et al., Reference Ulrich, Fernández, Jentzsch, Rolke, Schröter and Leuthold2006) involve the manipulation of different SOAs. Dual-tasks may vary in the requirements for the executing order of the two tasks, and a typical type of dual-task gives priority to Task 1. The processing mechanism of this type of dual-task is clearly elucidated in the response-selection bottleneck (RSB) model, as shown in Figure 1, adapted from Fischer and Plessow (Reference Fischer and Plessow2015) and Strobach et al. (Reference Strobach, Becker, Schubert and Kühn2015), and the same as Figure 1 in Zhong and Dong (Reference Dong, Derreira and Schwieter2023). The RSB model suggests a structural bottleneck curbing simultaneous processing of the central stages (e.g., “RSelect1” and “RSelect2” in Figure 1) of the two tasks, so “RSelect2” should be postponed till the end of “Rselect1,” resulting in prolonged Task 2 RT (Fischer & Plessow, Reference Fischer and Plessow2015). However, other researchers (see Mittelstädt et al., Reference Mittelstädt, Mackenzie, Leuthold and Miller2022; Ulrich et al., Reference Ulrich, Fernández, Jentzsch, Rolke, Schröter and Leuthold2006 for details) found that motor responses (“MotorR” in Figure 1) also to some extent contributed to the slowdown of Task 2 responses.
Figure 1. Illustration of a psychological refractory period dual-task with two stimulus onset asynchrony conditions for Task 2 (adapted from Fischer & Plessow, Reference Fischer and Plessow2015 and Strobach et al., Reference Strobach, Becker, Schubert and Kühn2015, and the same as the 1st figure in Zhong & Dong, Reference Dong, Derreira and Schwieter2023). The processing of each task proceeds from stimulus perception (Percp1/2) to response selection (RSelect1/2) and then to motor response (MotorR1/2). Response selection can only be processed serially, and RSelect2 can be processed only after the completion of RSelect1. TC1/TC2/TC3: task coordination at different stages.
The role of “RSelect” (response selection) and “MotorR” (motor response) in dual-task processing would be better explored by resorting to the ERP technique, especially using the ERP component of lateralized readiness potentials (LRPs). LRP can be either stimulus-locked or response-locked, with the former being associated with pre-motor processes like response selection (Huang & Luo, Reference Huang and Luo2006; Mittelstädt et al., Reference Mittelstädt, Mackenzie, Leuthold and Miller2022) and the latter, motor stages like motor response/execution (Mittelstädt et al., Reference Mittelstädt, Mackenzie, Leuthold and Miller2022). Besides, LRP also helps solve or minimize an inherent problem in investigating the neural responses of coordinating PRP dual-tasks with the ERP technique, i.e., the brain waves of the two tasks overlap with each other, and is hard to isolate either of them. According to Luck (Reference Luck2014), the process of subtracting the LRP waveform can exclude any brain responses elicited before the participants make a decision for behavioral responses. Specifically, when motor preparation occurs on one hand, the ERP wave is larger in the motor area of the opposite brain hemisphere than in that of the same side, and the difference wave shows the influence of the current task only, as at the time, the other task, whose preparation has finished or not started yet, elicits little hemispheric differences at this time window.
Osman and Moore (Reference Osman and Moore1993) verified the viability of using LRP to uncover the neural mechanism of processing a typical PRP task (with priority given to the first task). In the first experiment, Task 1 required participants to judge the pitch of a tone (high vs low) and, Task 2, the content of a letter (“L” vs “R”), with the SOA between the two tasks being 50 ms (short), 200 ms (medium), or 500 ms (long). For each task, participants had to respond to one stimulus (e.g., high tone) with a certain finger of the left hand, and to the other (e.g., low tone) with the same finger in the right hand. They obtained robust stimulus-locked and response-locked LRP components for each task, and the results showed that only stimulus-locked (but not response-locked) LRP onset latency for Task 2 (but not Task 1) was systematically influenced by SOA, exhibiting a typical PRP effect, i.e., the latency decreases as SOA increases. As the PRP effect reflects the effect of the structural bottleneck on the switch between the response-selection stages of the two tasks, the outcomes suggest that the stimulus-locked LRP onset latency can serve as an index for the bottleneck switching stage of dual-task processing. However, based on the discussions that motor responses may also contribute to dual-task interference (e.g., Mittelstädt et al., Reference Mittelstädt, Mackenzie, Leuthold and Miller2022; Ulrich et al., Reference Ulrich, Fernández, Jentzsch, Rolke, Schröter and Leuthold2006), the response-locked LRP onset latency would also be taken into account.
Coordination in a PRP dual-task is usually indexed by dual-task costs (performance differences between dual- and single-task conditions) in Task 1 or Task 2 or both, with smaller costs indicating better coordination skill (e.g., Becker et al., Reference Becker, Schubert, Strobach, Gallinat and Kühn2016; Liepelt et al., Reference Liepelt, Strobach, Frensch and Schubert2011; Zhong & Dong, Reference Dong, Derreira and Schwieter2023). Furthermore, this skill involves three sub-skills corresponding to three different processing stages, i.e., task instantiation at the start of dual-task processing (“task instantiation” hereafter), bottleneck access, and bottleneck switching, corresponding respectively to TC1, TC2, and TC3 in Figure 1 (Strobach et al., Reference Strobach, Becker, Schubert and Kühn2015). Better performance in both Task 1 and Task 2 would suggest an advantage in task instantiation or bottleneck access or both, and better performance restricted to Task 2 would suggest an advantage in bottleneck switching (Strobach et al., Reference Strobach, Becker, Schubert and Kühn2015). Therefore, the coordination advantage in Strobach et al. (Reference Strobach, Becker, Schubert and Kühn2015), as reviewed earlier, was related to task instantiation and/or bottleneck access, and that in Zhong and Dong (Reference Dong, Derreira and Schwieter2023), bottleneck switching.
The present study
As discussed earlier, the interpreter advantage related to the need for coordinating multiple processes, an essential property of interpreting, is underexplored, and little is known about its ERP neural correlates.
The present study thus aimed at enriching this highly underexplored field by using the ERP technique and a PRP dual-task to inquire into the neural correlates of the interpreter advantage in coordination. Dual-task costs would be used as indices for coordination, and the costs of LRP (both stimulus-locked and response-locked) onset latency would serve as the neural measures of the skill. Besides, the present study was intended to enrich the investigation into interpreting students since most of the extant literature on the interpreter advantage in coordination focused on professional interpreters (as mentioned earlier). Following Zhong and Dong (Reference Dong, Derreira and Schwieter2023), in which the coordination advantage only appeared at the intermediate (but not beginning) stage of interpreting training, the present study would recruit interpreters at the same stage (Chinese-English graduate students with around two years’ training), together with matched controls (in general L2 experience, working memory, age, intelligence, parents’ education level, etc.).
Coordination in interpreting is characterized by frequent switches between subtasks (e.g., listening and speaking in SI; listening and note-taking in CI), and thus the interpreter’s coordination advantage, if there is any, should be associated with the bottleneck switching sub-skill. Accordingly, we predicted that the interpreting students would exhibit smaller dual-task costs restricted to Task 2, consistent with the behavioral study of Zhong and Dong (Reference Dong, Derreira and Schwieter2023). As regards neural results, since response selection is the key processing stage in bottleneck switching, and stimulus-locked LRP is a reflection of response selection, only stimulus-locked LRP (not response-locked LRP) was expected to reveal this pattern of results (i.e., smaller dual-task costs restricted to Task 2).
Methods
Participants
As suggested by Boudewyn et al. (Reference Boudewyn, Luck, Farrens and Kappenman2018), 32 participants per group would be a good choice for between-group studies analyzing the LRP component. Given the availability of targeted participants and the potential need to exclude the data of some participants during preprocessing, we recruited altogether 80 Chinese-English bilingual graduate students between the ages of 22 and 26. All the participants were right-handed, as tested by Coren (Reference Coren1992), and had a normal or corrected-to-normal vision. They signed a written consent after learning the nature of the experiment and received monetary compensation after the experiment.
Participants’ background information, such as L2 age of acquisition and L2 learning history, was collected by an adapted version of the Language History Questionnaire 2.0 (Li et al., Reference Li, Zhang, Tsai and Puls2014). The amount of interpreting training was self-reported. Parents’ education level was measured by a seven-point scale (1: elementary school and below; 2: middle school; 3: high school; 4: junior college; 5: bachelor’s degree; 6: master’s degree; and 7: doctoral degree). Besides, participants’ L2 proficiency and intelligence were respectively tested by the Oxford Quick Placement Test (version 2, Syndicate, Reference Syndicate2001) and the Combined Raven’s Test (Department of Psychology, East China Normal University, 1991). Their working memory updating was gauged by a visual-spatial N-back task, and working memory span, an operation span task adapted from Unsworth et al. (Reference Unsworth, Heitz, Schrock and Engle2005).
Data from eight participants were excluded for two reasons. First, the accuracy rates for six participants were lower than 60% in the N-back task. Second, the electroencephalogram (EEG) data of two participants contained excessive artifacts (frequent ocular oscillation, drifting, etc.), resulting in few valid ERP epochs (less than three) in certain task conditions. Among the remaining 72 participants, 40 (12 male, 28 female), labeled as the Interpreting group, were first-year graduate students majoring in interpreting, and the other 32 (1 male, 31 female), labeled as the Control group, were first-year graduate students majoring in other branches of English, i.e., linguistics or literature. The interpreting group had received 2.21 years of interpreting training on average, and they were receiving on average 7.9 hours of interpreting training per week (4 hours of in-class training and on average 3.9 hours of after-class practice) in the semester in which they took part in the experiment. During the same period, the control group received no in-class training and seldom practiced interpreting (less than 0.1 hours per week on average), although they had received some interpreting training when they were undergraduate students. Table 1 summarizes the background information of these participants.
Table 1. Summary of participants’ background information (means with standard deviations (SDs) in brackets)
Notes: aIE (hours per week): hours of interpreting experience per week during the semester of the experiment; bAoA: age of acquisition; *** p < .001.
As shown in Table 1, the two participant groups differed in interpreting training experience (t = 10.564, p < .001, Cohen’s d = 2.24), but they were matched in other background information, including L2 experience other than interpreting experience (i.e., frequency of L2 use, L2 AOA, L2 learning history and L2 proficiency) and relevant personal traits (i.e., age, intelligence, parents’ education level, working memory updating, and working memory span, see Strobach et al. Reference Strobach, Becker, Schubert and Kühn2015; Zhong & Dong, Reference Dong, Derreira and Schwieter2023).
Experiment Task
Following Osman and Moore (Reference Osman and Moore1993) and Strobach et al. (Reference Strobach, Becker, Schubert and Kühn2015), the dual-task in the present study consisted of an auditory task (Task 1) requiring participants to determine the pitch of a tone that was either low (350 Hz) or high (3250 Hz), and a visual task (Task 2) requiring participants to determine the size of a square that was either small or large. They needed to press the keys “2,” “4,” “6,” and “8” on a small keypad with left index, left middle, right middle, and right index fingers, respectively. The four keys formed left-right (“4” and “6”) and top-down (“2” and “8”) response pairs, which are spatially different and easy to remember (Osman & Moore, Reference Osman and Moore1993). Besides, the key-stimulus linkage was counterbalanced across participants, given that different tones/squares corresponded to the same finger of different hands because as shown in Section 2.3 below, obtaining LRP for a certain task involves a procedure of averaging the wave elicited by the two hands.
The experiment always started with two single-task blocks of the auditory task in which only tones were presented, followed by blocks of the visual task in which only squares were presented. The fixed sequence of the single-task blocks aimed to assist participants in adapting to the rule of prioritizing the auditory task in dual-task blocks, and to remain consistent with previous studies (Becker et al., Reference Becker, Schubert, Strobach, Gallinat and Kühn2016; Strobach et al., Reference Strobach, Becker, Schubert and Kühn2015; Zhong & Dong, Reference Dong, Derreira and Schwieter2023) for better inter-study discussions. After the single-task blocks, eight dual-task blocks were presented, in which both auditory and visual stimuli were presented.
In single-task blocks, each trial began with a white fixation for 500 ms. Then, a tone was presented for 50 ms or a square was presented until a response was made or when it exceeded 1500 ms. In dual-task blocks, each trial began with a white fixation for 500 ms, followed by a tone of 50 ms, with a maximum RT of 1500 ms. Then, a square was presented after an SOA of 100 ms, 150 ms, or 450 ms, until a response was made or when it exceeded 1500 ms. There are two main reasons for this SOA manipulation. First, the manipulation may serve to reduce participants’ expectations for Task 2 onset, thus avoiding additional strategies in dual-task processing. Second, although not all studies utilizing the PRP paradigm implemented SOA manipulation, all the interpreter advantage studies did so, employing three distinct SOA levels (Becker et al., Reference Becker, Schubert, Strobach, Gallinat and Kühn2016; Strobach et al. Reference Strobach, Becker, Schubert and Kühn2015; Zhong & Dong, Reference Dong, Derreira and Schwieter2023). In addition, the inter-trial interval for both single- and dual-task blocks was 400 ms, 500 ms, or 600 ms (to reduce expectancy effect). The procedure of the trials in dual-task blocks is presented in Figure 2. The whole experiment was conducted against a gray background. Participants were asked to respond as accurately and quickly as possible and to respond to Task 1 first in the dual-task blocks. In addition, some “catch” trials with only tones presented were added into the dual-task blocks, to discourage participants from grouping responses (i.e., performing the two tasks as a couplet after they have made decisions for both tasks) (Osman and Moore, Reference Osman and Moore1993).
Figure 2. Schematic illustration of the trial procedure in the dual-task blocks.
Each single-task block consisted of 80 trials, with 40 for each type of stimuli (i.e., low/high tones or small/large squares). Each of the eight dual-task blocks consisted of 70 trials, with five trials for each type of stimulus pair (low/high tone ∼ small/large square) in each of the three SOA conditions, and 10 catch trials with five for each type of tone (low/high). In total, there were 80 trials for each stimulus type (low/high tone for Task 1 or small/large square for Task 2) in each condition (the single-task condition and the dual-task conditions of different SOAs). As different tones/squares were associated with different hands, there were 80 trials for each hand in each condition for each task, and thus the LRP in each condition for each task was assumed to be averaged across 80 trials (see Section 2.3 for details of obtaining LRP). In addition, there was a practice block of eight trials before the auditory single-task blocks and the visual single-task blocks, respectively, and a practice block of 14 trials before the dual-task blocks. The whole experiment lasted for around two hours, including the placement and removal of the caps for EEG recording.
EEG recording and offline processing
Online recording. Elastic electrode caps and Neuroscan 4.5 were used to continuously record the EEG data (sampling rate 2000 Hz, bandpass 0.05–500 Hz). The electrode caps have 64 Ag/AgCl electrodes whose arrangement was based on the International 10/20 system. The VEOG (vertical electrooculogram) was recorded by two electrodes, which were respectively placed above and below the left eye. The HEOG (horizontal electrooculogram) was recorded by two electrodes, which were placed at the outer canthi of both eyes. Online reference was set at the left mastoid. Impedances were kept below 10 KΩ.
Offline preprocessing. This procedure involved nine steps. (1) The sampling rate was reduced to 1000 Hz. (2) The EEG data were re-referenced to the average of M1 and M2 channels, following many previous LRP studies (e.g., Keye, et al., Reference Keye, Kinder, Rosok, Cannavale, Walk and Khan2024; Li et al., Reference Li, Liu, Zhang, Huang, Zhang, Liu and Chen2017; Morris et al., Reference Morris, Agirrezabal, Brännström and Gade2022; Osborne et al., Reference Osborne, Zhang, Farrens, Geiger, Kraus, Glazer and Mittal2022; Schmitt et al., Reference Schmitt, Münte and Kutas2000). (3) Trials with incorrect responses and artifacts such as drifting, myoelectricity, and HEOG were manually rejected. (4) VEOG was corrected by regression. (5) A digital low-pass filter (30 Hz, 12 dB/octave) was applied to the EEG data. (6) Epoching was divided into two parallel steps. First, to obtain the stimulus-locked LRP, the EEG data were segmented into epochs of 1200 ms, with a 200 ms pre-stimulus baseline. Second, to obtain the response-locked LRP, the filtered EEG data were segmented into epochs of 1750 ms, with a pre-response baseline of –1000 ms ∼ –750 ms. (7) The above epochs were baseline corrected. (8) Epochs with amplitude over ±70 μV were automatically rejected. The loss of epochs in all these steps was 15.53%. For stimulus-locked LRP, the mean number of accepted epochs in each condition was: 64.11 (±13.63) in Task 1 singe-task, 61.47 (±8.31) and 62.28 (±8.10) in Task 1 dual-task with SOA, respectively, at 100 ms and 150 ms, 67.88 (±12.26) in Task 2 singe-task, 61.29 (±8.19) and 62.06 (±8.35) in Task 2 dual-task with SOA, respectively, at 100 ms and at 150 ms. For response-locked LRP, the mean number of accepted epochs in each condition was: 65.04 (±13.01) in Task 1 singe-task, 61.81 (±8.34) and 62.83 (±8.14) in Task 1 dual-task with SOA, respectively, at 100 ms and 150 ms, 68.65 (±11.95) in Task 2 singe-task, 61.29 (±8.48) and 62.17 (±8.51) in Task 2 dual-task with SOA, respectively, at 100 ms and 150 ms. (9) Epochs of the same type (the same stimulus in the same single- or dual-task condition for the same task) were averaged, respectively.
Extracting the LRP waveform. The LRP waveform was extracted by subtracting the ipsilateral wave (average of the left hemisphere wave elicited by the left hand, and the right hemisphere wave elicited by the right hand) from the contralateral wave (average of the left hemisphere wave elicited by the right hand, and the right hemisphere wave elicited by the left hand) (Luck, Reference Luck2014). Following some previous studies (e.g., Boudewyn et al., Reference Boudewyn, Luck, Farrens and Kappenman2018; Gladwin et al., Reference Gladwin, M’t Hart and de Jong2008; Huang & Luo, Reference Huang and Luo2006; Zordan et al., Reference Zordan, Sarlo and Stablum2008), the left and right hemisphere waves (of the motor area) were represented by the waves in the electrodes C3 and C4, respectively.
LRP onset latency determination. The 50% peak latency measure was applied to determine the onset latency of the LRP component, which, according to Luck (Reference Luck2014), is the most reliable method in most cases. Following Luck (Reference Luck2014), LRP onset latency was determined as the time point at which the amplitude reaches half of the peak amplitude. Specifically, for each participant in each condition, we first detected the peak amplitude of the LRP wave and calculated its half value. Then we identified the largest time point (in terms of absolute value) before the peak, at which the amplitude of the LRP wave reached or started to exceed the half value.
Data analysis
Behavioral data. In the data trimming process, trials with no or incorrect responses (including responses with incorrect executing orders) and those with RT exceeding mean±3SD were excluded. In this procedure, 1.03% of the data were lost. Besides, some behavioral responses in the SOA = 450 ms condition were not successfully recorded, which may happen in a dual-task condition when participants have responded to Task 1 before the onset of Task 2, and as a result, the SOA = 450 ms condition did not enter further data analysis.
The remaining data consisted of four types, i.e., accuracy (ACC) and RT of Task 1 and those of Task 2. Dual-task costs of each type were separately subjected to the same ANOVA analysis with a between-group variable of Group (Interpreting, Control), and a within-group variable of Condition (dual-task_100, dual-task_150, with 100, 150 referring to different SOA conditions).
ERP data. These data consisted of four types, i.e., stimulus-locked and response-locked LRP of Task 1, and those of Task 2. Dual-task costs of each type were separately subjected to the same ANOVA analysis above.
In addition, three more analyses were conducted and the results were shown in the appendices. First, an analysis of raw data was also conducted and the results are presented in Appendix A. In this analysis, the raw data of ACC, RT, stimulus-locked LRP onset latency, and response-locked LRP onset latency of Task 1 and Task 2 were separately subjected to the same ANOVA analysis with a between-group variable of Group (Interpreting, Control), and a within-group variable of Condition (single-task, dual-task_100, dual-task_150, with 100, 150 referring to different SOA conditions). Second, participants may group their responses in dual-task processing (Ulrich & Miller, Reference Ulrich and Miller2008), and to minimize the impact of response grouping, which participants were asked to avoid but may have failed sometimes, we followed Osman and Moore (Reference Osman and Moore1993) and excluded 10% of the trials with the slowest Task 1 RT for each participant in each dual-task condition. The remaining data (both behavioral and ERP data) were subjected to the same data analysis procedure as mentioned earlier, and the results were shown in Appendix B. Briefly, the results pattern remained unchanged. Third, we also ran mixed-effects models for raw ACC and raw RT, and the results were presented in Appendix C. However, we did not apply mixed-effects models to the ERP data or dual-task costs of the behavioral data due to the nature of the data processing involved. Specifically, the averaging and subtracting procedure used in extracting the LRP component (as outlined in Section 2.3) and in calculating dual-task costs resulted in the absence of trial-based data, so the variance of the random effect was too small to be suitable for mixed-effects models.
Results
The results will be reported in the order of behavioral data (ACC, RT) and ERP data (stimulus-locked and response-locked LRP onset latency), with Task 1 being presented before Task 2 for each index.
Behavioral data
Table 2 lists the behavioral data.
Table 2. Group means (with SD) of raw data and dual-task costs in accuracy (ACC) (%) and in response time (RT) (ms) in each task and in each condition
Note: aSingle: single-task condition; bdual: dual-task, with 100 and 150 referring to the two stimulus onset asynchrony conditions; cdual-task cost: differences between dual- and single-task conditions.
ACC analysis of dual-task costs
Task 1. The main effect of Condition was significant (F(1, 70) = 10.81, p = .002, η p 2 = .134), i.e., dual-task costs were smaller in the dual-task_150 condition than in the dual-task_100 condition. The main effect of Group and its interaction with Condition were not significant (Fs < 3.1, p > .08).
Task 2. The main effect of Condition was significant (F(1, 70) = 13.45, p < .001, η p 2 = .161), i.e., dual-task costs were smaller in the dual-task_150 condition than in the dual-task_100 condition. The main effect of Group and its interaction with Condition were not significant (Fs < 4, p > .05).
RT analysis of dual-task costs
Task 1. The main effect of Condition was significant (F(1, 70) = 26.61, p < .001, η p 2 = .275), i.e., dual-task costs were smaller in the dual-task_150 condition than in the dual-task_100 condition. The main effect of Group and its interaction with Condition were not significant (Fs < 3.3, p > .07).
Task 2. The main effect of Group reached significance (F(1, 70) = 4.87, p = .031, η p 2 = .065), i.e., the Interpreting group exhibited smaller dual-task costs than the Control group, suggesting an interpreter advantage in coordination. The main effect of Condition was obtained (F(1, 70) = 398.12, p < .001, η p 2 = .850), i.e., the dual-task cost was smaller in the dual_150 condition than in the dual_100 condition. The interaction between Group and Condition was not significant (F(1, 70) = 2.83, p = .097, η p 2 = .039).
To summarize, the most important finding for behavioral data is that the Interpreting group exhibited smaller dual-task costs than the Control group only in Task 2 RT, suggesting an interpreter advantage in coordination, especially in bottleneck switching.
ERP data
Figure 3 depicts the grand-averaged LRP waves across different participant groups (Interpreting and Control), tasks (Task 1 and Task 2), and conditions (dual-task_100, dual-task_150, with 100, 150 referring to different SOA conditions).
Figure 3. Grand-averaged waves of each type of lateralized readiness potential (LRP) for the two participant groups in different tasks and in different conditions. (a) Waves for stimulus-locked LRP in Task 1, (b) waves for stimulus-locked LRP in Task 2, (c) waves for response-locked LRP of Task 1, and (d) waves for response-locked LRP in Task 2.
Stimulus-locked LRP onset latency analysis of dual-task costs
Task 1. None of the main effects or interactions reached significance (Fs < 2.2, p > .1).
Task 2. The main effect of Group was significant (F(1, 70) = 6.38, p = .014, η p 2 = .084), i.e., the dual-task costs for stimulus-locked LRP onset latency were smaller in the Interpreting group than in the Control group, suggesting an interpreter advantage in coordination. The main effect of Condition was also significant (F(1, 70) = 9.17, p = .003, η p 2 = .116), i.e., the dual-task costs were smaller in the dual-task_150 condition than in the dual-task_100 condition. The interaction between Group and Condition did not reach significance ((F(1, 70) = 1.19, p = .280, η p 2 = .017).
Response-locked LRP onset latency analysis of dual-task costs
Task 1. None of the main effects or interactions reached significance (Fs < 1, p > .09).
Task 2. None of the main effects or interactions reached significance (Fs < 3, p ≥ .1).
To summarize, the group difference in LRP onset latency, which is our primary concern, was found only in stimulus-locked LRP for Task 2, suggesting an interpreter advantage in the bottleneck switching sub-skill of coordination at the response preparation stage.
Discussion
The present study was the first study aiming at investigating the ERP neural correlates of the interpreter advantage in coordination. To achieve this goal, we recruited a group of interpreting students and a group of general bilingual controls to perform a PRP dual-task and analyzed the dual-task costs in their behavioral and neural responses. There were two main findings. First, the interpreting group exhibited smaller dual-task costs restricted to Task 2 in RT, suggesting an interpreter advantage in the bottleneck switching sub-skill of coordination. Second, the smaller dual-task costs restricted to Task 2 were also obtained in stimulus-locked LRP onset latency, offering neurological evidence for the bottleneck switching advantage. The first finding successfully replicated that of Experiment One in Zhong and Dong (Reference Dong, Derreira and Schwieter2023), while the second finding was novel, indicating the neural correlates of the interpreter advantage in coordination, i.e., smaller dual-task costs in the stimulus-locked LRP onset latency.
A unique contribution of the present study is that it delved into the time course of task processing with the ERP technique, and shed light upon the specific processing stage where the advantage lay. Theoretically, based on the RSB model, task processing can be broken down into three stages, i.e., (1) stimulus perception, (2) response selection, and (3) motor response, among which response selection can only be processed serially, and thus constitute a bottleneck in the processing of dual-tasks (see Figure 1 for details). Therefore, the essence of bottleneck switching can be taken as the switch between response-selection stages of Task 1 and Task 2. In other words, an advantage in bottleneck switching should be manifested in an advantage in response selection, which was verified in the ERP results of the present study (interpreting students’ better performance in stimulus-locked LRP).
The presence of an interpreter advantage in the stimulus-locked LRP, together with the absence of that advantage in the response-locked LRP, suggests that interpreters outperformed general bilinguals in choosing an appropriate action instead of executing the action in the coordination of two simultaneously presented tasks. This advantage in coordination is consistent with what is required in coordinating multiple tasks in consecutive interpreting. Consecutive interpreters have to intermittently switch between listening, note-taking, memorizing, etc. in the listening phase, and speaking, note-reading, recalling, etc. in the speaking phase, and to efficiently switch from the former phase to the latter. These results may therefore expound the essence of multitask coordination in consecutive interpreting, i.e., efficient switching between the response-selection stages of different subtasks of an interpreting task.
The nature of the advantage found in the present study is consistent with that in Zhong and Dong (Reference Dong, Derreira and Schwieter2023), but different from that in Strobach et al. (Reference Strobach, Becker, Schubert and Kühn2015). Both the present study and Zhong and Dong (Reference Dong, Derreira and Schwieter2023) found that interpreters’ better performance was restricted to Task 2, indicating an advantage in the bottleneck switching sub-skill of the coordination skill, while Strobach et al. (Reference Strobach, Becker, Schubert and Kühn2015) found interpreters’ better performance in both Task 1 and Task 2, exhibiting an advantage in task instantiation, or bottleneck access or both of them. The consistency and inconsistency of the above findings are worth in-depth discussion since the three studies employed quite similar PRP tasks, i.e., an auditory task (pitch discrimination) as Task 1, and a visual task (size discrimination) as Task 2, with three different SOAs. As explained by Zhong and Dong (Reference Dong, Derreira and Schwieter2023), the above difference is relevant to the type of participants recruited, i.e., consecutive interpreting students (the present study; Zhong & Dong, Reference Dong, Derreira and Schwieter2023) vs. professional simultaneous interpreters (Strobach et al., Reference Strobach, Becker, Schubert and Kühn2015). Consecutive interpreting students are required to efficiently switch between different phases, and between different subtasks in each phase, but professional simultaneous interpreters are probably more experienced in taking global control of the whole task (Zhong & Dong, Reference Dong, Derreira and Schwieter2023). While we have not directly compared the present study with Becker et al. (Reference Becker, Schubert, Strobach, Gallinat and Kühn2016), which employed the same participant sample as Strobach et al. (Reference Strobach, Becker, Schubert and Kühn2015), it is worth noting that Becker et al. (Reference Becker, Schubert, Strobach, Gallinat and Kühn2016) focused primarily on fMRI data and reported behavioral results only for Task 2. This differs from our study, which reported behavioral results of both Task 1 and Task 2, together with the corresponding ERP data.
Besides, our results differ from Padilla et al. (Reference Padilla, Bajo and Macizo2005) and Morales et al. (Reference Morales, Padilla, Gómez-Ariza and Bajo2015) who did not report an interpreter advantage in coordination. As mentioned in “Introduction,” a possible explanation for this discrepancy lies in the different measures employed in those studies. Specifically, Padilla et al. (Reference Padilla, Bajo and Macizo2005) used non-speeded responses, while Morales et al. (Reference Morales, Padilla, Gómez-Ariza and Bajo2015) relied on accuracy. According to Strobach et al. (Reference Strobach, Becker, Schubert and Kühn2015), these measures may not be as sensitive as the PRP dual-task paradigm with RT indices, which we employed in the present study.
It is important to note that the RSB model, on which the theoretical reasoning of the present study is based, is not the sole model for explaining PRP dual-task processing. For example, the resource-sharing model posits that the dual-task interference is associated with the allocation of the divisible resources to the two tasks, whose central stages can be processed in parallel (see Navon & Miller, Reference Navon and Miller2002 for detailed discussions). Future studies based on other dual-task processing models will definitely contribute to a thorough understanding of the mechanism underlying the interpreter advantage in coordination.
To conclude, the present study not only replicated the interpreter advantage in the bottleneck switching sub-skill of coordination in Zhong and Dong (Reference Dong, Derreira and Schwieter2023) but also found the neural correlates of the advantage, i.e., smaller dual-task costs restricted to Task 2 in the stimulus-locked LRP onset latency. Though the issue of the interpreter advantage in coordination is made clearer, more issues await further investigation, such as the neural correlates of interpreter advantages in professional interpreters, the developmental changes of the neural correlates across different interpreting training stages, etc.
Acknowledgments
This work was supported by the National Social Science Found of China under grant [22BYY077]. We thank Xiaochong Chen, Zhibin Yu, Hongming Zhao, Ge Xu, Jiadan Lin, Hao Wen, and Wenzhen Zhu for their help with the research.
Competing interests
The authors declare no competing interests.
Open Practice Statement
The data that support the findings of this study are openly available in Mendeley Data at http://doi.org/10.17632/x92jnbnj7c.1
Appendix A. Results of Raw Data Analysis
As mentioned in Section 2.4, this appendix reports the results of raw data analysis, in which better coordination is indexed by an interaction between Group and Condition, with one group showing better performance in the dual-task condition(s) than the other, but not in the single-task condition (c.f., Strobach et al., Reference Strobach, Becker, Schubert and Kühn2015). Therefore, the raw data of ACC, RT, stimulus-locked LRP onset latency, and response-locked onset latency of Task 1 and Task 2 were separately subjected to the same ANOVA analysis with a between-subject variable of Group (Interpreting, Control), and a within-subject variable of Condition (single-task, dual-task_100, dual-task_150, with 100, 150 referring to different SOA conditions). If an interaction between Group and Condition was obtained, further analysis would focus on group differences in each condition.
Similar to the report in the main text, the results of ACC, RT, stimulus-locked, and response-locked LRP onset latency will be reported in this order, with Task 1 being presented before Task 2 for each index.
Behavioral data: ACC analysis of the raw data
Task 1. The main effect of Condition was significant (F(2, 140) = 13.28, p <. 001, η p 2 = .159), with Task 1 ACC being significantly decreasing in the following order: single-task condition, dual-task_150 condition, and the dual-task_100 condition (p < .05). The main effect of Group and its interaction with Condition did not reach significance (Fs < 2.5, p > .1).
Task 2. The main effect of Condition was significant (F(2, 140) = 33.44, p <. 001, η p 2 = .323), with Task 2 ACC being significantly decreasing in the following order: single-task condition, dual-task_150 condition, and the dual-task_100 condition (p ≤ .001). The main effect of Group and its interaction with Condition did not reach significance (Fs < 4, p > .05).
In short, although the main effect of Condition was observed, no significant group differences were obtained for the ACC data, which is our major concern. Group means and standard deviations are summarized in Table 2 in the main text.
Behavioral data: RT analysis of the raw data
Task 1. The main effect of Condition was significant (F(2, 140) = 313.82, p <. 001, η p 2 = .818), with Task 1 RT significantly increasing in the following order: single-task condition, dual-task_150, and dual-task_100 conditions (all p < .001). The main effect of Group and its interaction with Condition did not reach significance (Fs < 2.8, p > .08).
Task 2. The main effect of Group reached significance (F(1, 70) = 5.91, p = .027, η p 2 = .068), with the Interpreting group exhibiting faster responses to Task 2 than the Control group. The interaction between Group and Condition was also significant (F(2, 140) = 4.77, p = .029, η p 2 = .064). Simple effect analysis showed that the Interpreting group revealed faster responses to Task 2 in the dual-task_150 (p = .036) and the dual-task_100 (p = .017) conditions, but not in the single-task condition (p = .205), suggesting an interpreter advantage in coordination. The main effect of Condition was significant (F(2, 140) = 924.39, p <. 001, η p 2 = .930), with Task 2 RT significantly increasing in the following order: single-task condition, dual-task_150 condition, and dual-task_100 conditions (all p < .001).
To be short, the Interpreting group exhibited faster responses only in the dual-task condition, but not in the single-task condition, suggesting an interpreter advantage in coordination, and the above results were only observed in Task 2, but not in Task 1, suggesting an interpreter advantage in the bottleneck switching sub-skill of coordination.
ERP data: Stimulus-locked LRP onset latency analysis of the raw data
Task 1. The main effect of Condition was significant (F(2, 140) = 55.14, p <. 001, η p 2 = .441), i.e., stimulus-locked LRP onset latency was shorter in the single-task condition than in the two dual-task conditions (p < .001), with the dual-task conditions not differing from each other (p = 1.000). The main effect of Group and its interaction with Condition did not reach significance (Fs < 2, p > .1).
Task 2. The main effect of Group reached significance (F(1, 70) = 5.38, p =. 023, η p 2 = .071), with the Interpreting group exhibiting shorter stimulus-locked LRP onset latency in Task 2 than the Control group. The interaction between Group and Condition was also significant (F(2, 140) = 4.12, p = .018, η p 2 = .056). Simple effect analysis showed that the Interpreting group revealed shorter stimulus-locked LRP onset latency in Task 2 in the dual-task_150 (p = .002) condition, but not in the single-task condition (p = .667) and dual-task_100 (p = .182) condition, suggesting an interpreter advantage in coordination. The main effect of Condition was significant (F(2, 140) = 211.20, p <. 001, η p 2 = .751), with Task 2 stimulus-locked LRP onset latency significantly increasing in the following order: single-task condition, dual-task_150 condition, and dual-task_100 conditions (all p ≤ .012).
To summarize, the Interpreting group exhibited shorter stimulus-locked LRP onset latency only in the dual-task condition, but not in the single-task condition, suggesting an interpreter advantage in coordination, and the above results were only observed in Task 2, but not in Task 1, suggesting an interpreter advantage in the bottleneck switching sub-skill of coordination.
There was a slight discrepancy between the results of the raw data and those of dual-task costs. That is, analysis of the raw data did not reveal group differences in the dual-task_100 condition (though it did in the dual-task_150 condition), while analysis of dual-task costs did (with a main effect of Group and no interaction between Group and Condition). A closer look at the raw data reveals that the stimulus-locked LRP onset latency for the Interpreting group was numerically, though not significantly, larger in the single-task condition and smaller in the dual-task_100 condition, compared with the Control group. In this circumstance, the group difference in the dual-task cost (dual-minus-single) equaled the addition of the group differences in the single-task condition and the dual-task_100 condition (of the raw data), which is why the group difference reached significance in the dual-task costs although it did not in the raw data. In other words, dual-task cost seems more sensitive to group differences than the raw data. Given this, we would not take the insignificant group difference in the single-task and the dual-task_100 conditions (with significant group differences in the dual-task cost) as indicating no interpreter advantage in the dual-task_100 condition.
ERP data: Response-locked LRP onset latency analysis of the raw data
Task 1. The main effect of Condition was significant (F(2, 140) = 33.33, p <. 001, η p 2 = .323), i.e., response-locked LRP onset latency was shorter in the single-task condition than in the two dual-task conditions (p < .001), with the dual-task conditions not differing from each other (p = 1.000). The main effect of Group and its interaction with Condition did not reach significance (Fs < 1, p > .1).
Task 2. The main effect of Condition was significant (F(2, 140) = 17.88, p <. 001, η p 2 = .203), i.e., response-locked LRP onset latency was shorter in the single-task condition than in the two dual-task conditions (p < .001), with the dual-task conditions not differing from each other (p = 1.000). The main effect of Group and its interaction with Condition did not reach significance (Fs < 2, p > .1).
In a word, although the main effect of Condition was observed, no significant group differences were obtained in the response-locked LRP onset latency.
Appendix B. Results excluding 10% of the dual-task trials
As mentioned in Section 2.4 (“Data analysis”), to minimize the impact of response grouping, we followed Osman and Moore (Reference Osman and Moore1993) and excluded 10% of the trials with the slowest Task 1 RT for each participant in each dual-task condition. The remaining data (both behavioral and ERP data) were subjected to the same data analysis procedure as mentioned in the main text, i.e., dual-task costs were subjected to the ANOVA analysis with Group (Interpreting, Control) as the between-group variable, and Condition (dual-task_100, dual-task_150, with 100, 150 referring to different SOA conditions) as the within-group variable. The results are presented below. Briefly, the results pattern remained unchanged.
Table B1 lists the behavioral data.
Table B1. Group means (with SD) of raw data and dual-task costs in accuracy (ACC) (%) and in response time (RT) (ms) in each task and in each condition
Note: asingle: single-task condition; bdual: dual-task, with 100 and 150 referring to the two stimulus onset asynchrony conditions).
Behavioral data: ACC analysis of dual-task costs
Task 1. The main effect of Condition was significant (F(1, 70) = 10.81, p = .002, η p 2 = .134), i.e., dual-task costs were smaller in the dual-task_150 condition than in the dual-task_100 condition. The main effect of Group and its interaction with Condition were not significant (Fs < 3.1, p > .08).
Task 2. The main effect of Condition was significant (F(1, 70) = 13.45, p < .001, η p 2 = .161), i.e., dual-task costs were smaller in the dual-task_150 condition than in the dual-task_100 condition. The main effect of Group and its interaction with condition were not significant (Fs < 4, p > .05).
Behavioral data: RT analysis of dual-task costs
Task 1. The main effect of Condition was significant (F(1, 70) = 19.31, p < .001, η p 2 = .216), i.e., dual-task costs were smaller in the dual-task_150 condition than in the dual-task_100 condition. The main effect of Group and its interaction with Condition were not significant (Fs < 3, p > .1).
Task 2. The main effect of Group reached significance (F(1, 70) = 4.79, p = .032, η p 2 = .064), i.e., the Interpreting group exhibited smaller dual-task costs than the Control group, suggesting an interpreter advantage in coordination. The main effect of Condition was obtained (F(1, 70) = 360.23, p < .001, η p 2 = .837), i.e., the dual-task cost was smaller in the dual_150 condition than in the dual_100 condition. The interaction between Group and Condition was not significant (F(1, 70) = 2.53, p = .116, η p 2 = .035).
To summarize, the most important finding for behavioral data is that the Interpreting group exhibited smaller dual-task costs than the Control group in and only in Task 2 RT, suggesting an interpreter advantage in coordination, especially in bottleneck switching.
ERP data: Stimulus-locked LRP onset latency analysis of dual-task costs
Task 1. None of the main effects or interactions reached significance (Fs < 2, p > .1).
Task 2. The main effect of Group was significant (F(1, 70) = 4.67, p = .034, η p 2 = .063), i.e., the dual-task costs for stimulus-locked LRP onset latency was smaller in the Interpreting group than in the Control group, suggesting an interpreter advantage in coordination. The main effect of Condition was also significant (F(1, 70) = 7.48, p = .008, η p 2 = .096), i.e., the dual-task costs were smaller in the dual-task_150 condition than in the dual-task_100 condition. The interaction between Group and Condition did not reach significance ((F(1, 70) = .033, p = .856, η p 2 < .001).
ERP data: Response-locked LRP onset latency analysis of dual-task costs
Task 1. None of the main effects or interactions reached significance (Fs < 2.5, p > .1).
Task 2. None of the main effects or interactions reached significance (Fs < 2, p > .1).
To summarize, the group difference in LRP onset latency, which is our primary concern, was found only in stimulus-locked LRP for Task 2, suggesting an interpreter advantage in the bottleneck switching sub-skill of coordination at the response preparation stage.
Appendix C. Results of mixed-effects models
As mentioned in Section 2.4, Appendix C reports the results of mixed-effects models. Mixed-effects models were run for the raw behavioral data only, instead of behavioral dual-task costs and ERP data, as the latter two did not yield trial-based data, rendering the variance of random effects too small to be suitable for mixed-effects models.
Data analysis was conducted using the R package lme4 (Bates, Maechler, Bolker, & Walker, Reference Bates, Maechler, Bolker and Walker2014). Models for ACC data were fit with the function glmer, and models for RT data were fit with the function lmer. The package lmerTest (Kuznetsova et al., Reference Kuznetsova, Brockhoff and Christensen2017) was used to estimate p values, and the package emmeans (Lenth et al., Reference Lenth, Singmann, Love, Buerkner and Herve2018) was used for pairwise comparisons. Below are the model formulas for analyzing the ACC data and the RT data, respectively:
ACC data: glmer(ACC ∼ Group * Condition + (1|Participant), family=binomial)
RT data: lmer(RT ∼ Group * Condition + (1|Participant))
where Group∈(Interpreting, Control); Condition∈(single-task, dual-task_100, dual-task_150, with 100, 150 referring to different SOA conditions)
Table C1 lists the results of the ACC data, and Table C2 lists those of the RT data. The results of the mixed-effects models are reported below.
Table C1. Results of generalized linear mixed-effects models for accuracy data
Note: asingle: single-task condition; bdual: dual-task, with 100 and 150 referring to the two stimulus onset asynchrony conditions).
Table C2. Results of generalized linear mixed-effects models for response time data
Note: asingle: single-task condition; bdual: dual-task, with 100 and 150 referring to the two stimulus onset asynchrony conditions.
Task 1 ACC. As shown in Table C1, no group difference was obtained for overall ACC (b = .03, p = .715), but Group interacted with both the difference between the dual-task_100 and single-task conditions (b = −.34, p = .033) and that between the dual-task_150 and single-task conditions (b = −.46, p = .005). Pairwise comparisons showed that the two groups did not differ significantly in the single-task and dual-task_100 conditions (p > .05), but the Interpreting group exhibited larger ACC than the Control group in the dual-task_150 conditions (p = .035). Besides, the accuracy in the single-task condition was larger than in the dual-task_100 condition (p < .001), but not in the dual-task_150 condition (p = .199).
Task 2 ACC. As shown in Table C1, no group difference was obtained for overall ACC (b = .07, p = .908), but Group interacted with both the difference between the dual-task_100 and single-task conditions (b = .45, p = .002) and between the dual-task_150 and single-task conditions (b = .42, p = .005). Pairwise comparisons showed that the two groups did not differ significantly in the single-task and dual-task_150 conditions (p > .05), but the Interpreting group exhibited larger ACC than the Control group in the dual-task_150 conditions (p = .040). Besides, the accuracy in the single-task condition was larger than in the dual-task_100 and dual-task_150 conditions (p < .001).
Task 1 RT. As shown in Table C2, no group difference was obtained for overall RT (b = - 6.11, p = .796), but Group interacted with both the difference between the dual-task_100 and single-task conditions (b = 31.75, p < .001), and between the dual-task_150 and single-task conditions (b = 28.59, p < .001). Pairwise comparisons showed that the two groups did not differ significantly in any of the conditions (p > .05). Besides, the RT in the single-task condition was smaller than in the dual-task_100 and the dual-task_150 conditions (p < .001).
Task 2 RT. As shown in Table C2, no group difference was obtained for overall RT (b = 14.72, p = .507), but Group interacted with both the difference between the dual-task_100 and single-task conditions (b = 57.39, p < .001), and between the dual-task_150 and single-task conditions (b = 47.25, p < .001). Pairwise comparisons showed that the two groups did not differ significantly in the single-task condition (b = - 14.7, p = .504), but the Interpreting group exhibited smaller RT than the Control group in the dual-task_100 (b = - 72.1, p = 001) and dual-task_150 conditions (b = - 62.0, p = .005), suggesting an interpreter advantage in coordination. Besides, the RT in the single-task condition was smaller than in the dual-task_100 and dual-task_150 conditions (p < .001).
To be short, the Interpreting group exhibited faster responses only in the dual-task condition, but not in the single-task condition, suggesting an interpreter advantage in coordination, and the above results were only observed in Task 2, but not in Task 1, suggesting an interpreter advantage in the bottleneck switching sub-skill of coordination.
Although the pattern of RT results in the mixed-effects models was exactly the same as that obtained from the ANOVA (as reported in the main text), there were subtle discrepancies in the ACC outcomes. The ANOVA revealed no significant group differences, but mixed-effects models exhibited larger ACC for the Interpreting group than for the Control group in the dual-task_150 condition in Task 1, and in the dual-task_100 condition in Task 2, with no group differences in single-task conditions for both Task 1 and Task 2, suggesting better coordination skill for the Interpreting group. However, we refrain from attributing these results to any of the three components of the coordination skill, as the discussion on the interpreter advantage in these components was primarily based on processing efficiency rather than accuracy, and the experimental conditions that yielded significant group differences in ACC were not consistent across different tasks.
