Testing the efforts model of simultaneous interpreting: An ERP study

Roman Koshkin; Yury Shtyrov; Andriy Myachykov; Alex Ossadtchi

doi:10.1371/journal.pone.0206129

Abstract

We utilized the event-related potential (ERP) technique to study neural activity associated with different levels of working memory (WM) load during simultaneous interpretation (SI) of continuous prose. The amplitude of N1 and P1 components elicited by task-irrelevant tone probes was significantly modulated as a function of WM load but not the direction of interpretation. Furthermore, the latency of the P1 increased significantly with WM load. The WM load effect on N1 latency, however, did not reach significance. Larger negativity under lower WM loads suggests that more attention is available to process the source message, providing the first electrophysiological evidence in support of the Efforts Model of SI. Relationships between the direction of interpretation and median WM load are also discussed.

Citation: Koshkin R, Shtyrov Y, Myachykov A, Ossadtchi A (2018) Testing the efforts model of simultaneous interpreting: An ERP study. PLoS ONE 13(10): e0206129. https://doi.org/10.1371/journal.pone.0206129

Editor: Stefan Elmer, Universitat Zurich, SWITZERLAND

Received: March 20, 2018; Accepted: October 8, 2018; Published: October 24, 2018

Copyright: © 2018 Koshkin et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: The pre-processed dataset is available at the following URL: http://bit.ly/2FV4m57.

Funding: This work was supported by the Center for Bioelectric Interfaces of the Institute for Cognitive Neuroscience of the National Research University Higher School of Economics, RF Government grant, ag. No. 14.641.31.0003. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Introduction

Unlike in monolingual communication, in simultaneous interpreting (SI) a message in one language is perceived and processed almost concurrently with the production of an equivalent message in another language. To be able to accomplish this feat, besides high proficiency in both the source and target languages, the interpreter must possess a set of specialized skills, including exceptional language switching abilities [1], large working memory (WM) span [2], ability to manipulate WM content and understand incoming discourse while producing a rendering of an earlier portion of the source message in the target language. By its nature, SI is externally paced [3,4], indicating the need for cognitive resource management and coping strategies [5–10].

In SI, an interpreter usually begins interpreting before the speaker has finished a sentence. The speaker, however, does not normally wait to move on to the next utterance, regardless of whether the interpreter has completed the translation of the previous chunk [5,11]. Moreover, it may not always be possible or convenient to maintain sequential linearity of the target message relative to the source. For example, interpreters often reverse the order of lists. In some language combinations, e.g. German/English, syntactic constraints force one to wait for the final verb in the German source to construct the target sentence in English [12]. Finally, the interpreter may choose to defer translating a word until a good enough equivalent comes to mind, hoping to be able to work it into the target message later. The resulting source-target lag—also referred to as décalage or ear-voice-span (EVS) in the interpretation studies literature—between the source and the target messages highlights the critical role of WM in the SI pipeline. WM represents a mental space within which to perform the transformations needed for a coherent and accurate target message to emerge.

Under normal circumstances, when the source message is relatively easy to understand and target equivalents are quickly and automatically retrieved from long-term memory (LTM), the interpreter maintains a comfortable décalage, accurately rendering the source message with almost no omissions. But when confronted with a long-winded, dense or obscure passage, the interpreter may be forced out of the comfort zone and temporarily increase the lag to accommodate the need for more time to process it. The lag is similar to debt in that beyond a certain point it becomes difficult to handle. In extreme cases, when the interpreter gets too far behind to speaker, performance quality may be compromised: parts of the source message may get severely distorted or go missing from the translation altogether. This may happen when the interpreter has shifted much of his/her attention away from the currently articulated source chunk to finish processing the previous one stored in WM, in order to catch up with the speaker. In sum, large lags are most likely caused by processing difficulties.

On the other hand, when the source message is overall relatively difficult to follow (e.g. when the message is not in the interpreter’s mother tongue), the interpreter may need to allocate extra effort towards understanding. This can be done by shortening the décalage, effectively limiting the amount of information to be processed in working memory. Such a strategy may result in a more literal translation that is likely to be syntactically and grammatically deficient.

In our opinion, the above considerations are best captured by Gile’s Efforts Model [13] which conceptualizes SI in terms of three groups or mental operations, or ‘efforts’: listening, production and memory. Since these efforts are mostly non-automatic and concurrent, they critically depend on and compete for the limited pool of attentional resources. A major implication of the model is that increased processing demands in one of the efforts can only be met at the expense of another. In fact, several studies involving dual-task situations indirectly support this view suggesting that transient performance decreases in one task occur due to the engagement of attention in another task (e.g. [14]).

To our knowledge, only one study has attempted to test the Efforts Model of SI experimentally [15]. But as its author himself admitted, “it cannot be said to have led to [its] systematic testing or validation” and also suggested that “precise quantitative measurement” would help to make it more useful. To address this concern (at least partially), in the present paper we used the ERP technique to test one particular prediction of the Efforts Model, namely that increased processing demands on the ‘memory effort’ means less processing capacity available to the ‘listening effort’ (that involves active processing of the input heard). In other words, a higher WM load would create a deficit of attention to the auditory stream. Whereas this hypothesis may seem quite intuitive, to our knowledge, it has never been tested experimentally in a naturalistic setting requiring the participants to interpret continuous prose overtly. Electrophysiological evidence supporting it would suggest that interpreters’ brains gate part of the auditory input to be able to properly process the information backlog and reduce the associated processing pressure. Here and throughout we refer to the ‘memory’ and ‘listening’ efforts as defined by Gile [13].

We exploited the previous findings that N1 and even P1 [16] amplitude evoked by task-irrelevant probes embedded in a speech stream is modulated by selective attention in what is called ‘processing negativity' observed as early as 50–150 ms from the stimulus onset [17]. Specifically, the ERP waveform appears shifted towards negative values when the listener attends the target audio (e.g. [18–23]). Moreover, a more recent EEG study [24] showed that in a multitasking situation—and SI is an extreme case of multitasking [25]—increased WM load decreases attention to the targets. Therefore, the parameters of these early auditory ERP components can be used as a suitable and temporally precise index of interpreters’ attention to the spoken source message.

Our assumption that WM overload reduces attention to the auditory stream, which in turn modulates the ERP waveform, aligns well with the evidence that both WM and attention may utilize a common pool of neural resources [26]. As demonstrated by fMRI studies, attention and WM are subserved by overlapping brain areas [27–33].

While according to some studies–notably, not involving the Russian-English language pair used here–interpreters may prefer to interpret from L2 into L1 (e.g. [34,35] but see [36]), in our survey conducted prior to this study, out of 32 professional simultaneous interpreters (English-Russian/Russian-English, L1 Russian speakers), 29 reported that, all else being equal, interpreting from L2 into L1 was much more difficult than in the opposite direction. Furthermore, the interpreters we surveyed also said the most difficult part for them was to understand the source message in L2—and understanding is part of the ‘listening effort’ according to Gile [13,15,37]. Based on these observations and the prediction of the Efforts Model we expected that this subjective difficulty would result in a significant difference in median WM loads between L1→L2 and L2→L1 directions of interpretation.

Being aware that the early P1 and N1 ERP components produced by task-irrelevant tone probes are not known to be sensitive to the language of the speech that these probes are embedded into, we nevertheless wished to check whether the direction of interpretation would have an indirect effect on the amplitude of these early ERP components. The rationale for this was as follows. Given that bilinguals demonstrate superior memory performance for L1 words [38,39], we assumed that for L1 Russian/L2 English interpreters English→Russian interpreting involves a greater memory effort than Russian→English interpreting. If the Efforts Model is correct in predicting that larger memory effort means less attention to the current auditory input, we can hypothesize that increasing the lag (and by assumption WM load) by N words in the L2→L1 direction of interpreting should decrease P1/N1 negativity by a larger amount than when the lag is increased by the same number of words in L1→L2 interpreting.

Importantly, achieving the goals of the study required defining a method of estimating WM load. The most obvious and straightforward approach to do it would be to simply assume that WM load is proportional to the number of words that an interpreter lags behind the source message at any given time. In the time intervening between the moment of hearing a word and finishing its translation overtly, the interpreter has to store it in WM at least until it is further processed as appropriate to the situational context. If we assume that the cognitive effort it takes to process the most frequent function words—especially articles and prepositions—is negligible, we can exclude them from WM load estimation. However, this approach would be an oversimplification. First, it is predicated on the assumption that each word strains WM capacity equally. More frequent words must have stronger mental representations and a larger network of associations than relatively rare ones. In what is known as the word frequency effect [40–42] more frequent words are processed faster. Therefore, they must be translated more easily and cleared from working memory relatively quickly to make room for processing the next chunk of the source message. Conversely, words that are less familiar take more time and effort to process, which delays the retrieval of equivalents from LTM and increases WM load. Second, such an approach does not take into account that words in the source message are never chunked together such that they are stored in WM as a single mental object. Third, it assumes that WM load is reduced at the moment when a word has been translated overtly. This is not always the case because interpreters continue to post-process their own translation, in a phenomenon known as self-monitoring [37,43–46]. This means that after a word (or, more generally, a chunk) is translated overtly, it is not cleared from WM immediately, but at a later time, when the interpreter is satisfied with his or her translation. An opposite situation may happen—and this is the fourth reason—when the interpreter can use informational redundancy of the source message to predict an idea before it is fully uttered by the speaker [5]. For example, if the source message has multiple references to the United Nations Organization, it is easy to finish the translation before the offset of the phrase the United … [Nations Organization], in other words get ahead of the speaker. The above considerations mean that WM load estimates will be inherently noisy. To mitigate this error of measurement, we sought to maximize the sample and explored several alternative methods of WM load estimation, which we describe below.

Methods

Participants

Nine males (aged 25–47, M = 36.9, SD = 6.757) participated in the study. All were qualified interpreters holding a university degree in translation and interpreting, L1 Russian/L2 English speakers, with an average of 10.65 (SD = 6.675) years of professional SI experience. None of them reported having any hearing-related problems, substance abuse or a neurological disorder. The experiment was approved by the Institutional Review Board of the National Research University Higher School of Economics, Moscow. All the participants signed an informed consent form.

Procedure and materials

Each of the participants was asked to interpret 8 speeches originally delivered at the 6849th United Nations Security Council Meeting by representatives of eight countries: Chile, Colombia, Costa Rica, Honduras, Peru, Uruguay (in Spanish) and France and Morocco (in French). Since we were interested in the Russian-English language combination, we used 4 Russian translations and 4 English translation (see S1 and S2 Appendices for details) available on the official UN website at http://www.un.org. The topics discussed in the speeches centered on the rule of law and were unlikely to elicit a strong emotional response. We deliberately chose speeches originally delivered in a language other than the participants’ L1 or L2. This was necessary to avoid a potential bias due to some subtle properties (e.g. idiosyncratic syntax) that may have been present in the translated, but not in the original texts, and vice versa.

We expected that UN terminology would be familiar enough to the participants so that they could deliver quality translation with no preparation. To control for a potentially confounding effect of varying delivery speed, we had the written Russian and English translations read at a slow constant rate by a bilingual speaker (female) highly proficient in both Russian and English. After the recording we used Audacity (http://audacityteam.org), open audio editing software, to edit the recordings to a constant delivery rate of 105 words per minute (wpm). Thus we eliminated possible confounds that could have been introduced due to individual speaker’s voice features such as rate, pitch, timbre, loudness, prosody and accent. The total playback time was about 54 minutes (excluding periods of rest between the speeches).

The source speech recordings and task-irrelevant probe stimuli (440-Hz 52-ms pure sine wave tones including a rise and fall period of 4 ms) were played by a custom script running under PsychoPy [47]. The specific choice of 52 ms was a tradeoff between the need to maximize the amplitude of the expected ERPs (by longer probes) and minimize the masking of the source audio by too long probes. The probes were delivered throughout the entire performance of the interpreting task simultaneously to both the right and left ear with a jittered inter-stimulus interval (ISI) of 450–750 ms (M = 600 ms). These parameters were selected empirically to maximize the number of probes per second of experimental time while minimizing the effect of diminished ERP amplitude with shorter ISIs [21].

To control for order effects, the speeches were delivered to the participants in a pseudo-random fashion according to a Latin square design such that for every participant the texts’ order was different.

The speech recordings were played through earbud headphones (Sennheiser MX 170, 32Ω, Germany). Before the experiment, the participants did two 30-second practice runs, in which they were asked to adjust the volume to a comfortable level and interpret an excerpt from a speech delivered at the same UNSC session, but not included in the experimental material.

Participants were seated comfortably in a reclining chair in an electrically and acoustically shielded room. To reduce oculo-motor and muscle artifacts, they were instructed to sit still, relax, minimize eye movements and articulate their translation as quietly as possible. The translation was recorded using a Boya BY-M1 capacitor microphone for subsequent offline transcription and timecoding.

Working memory load estimation

The number of content words in the source-target lag provides a measure of instantaneous WM load. However, a potentially more precise estimate could be achieved by scaling every content word by its min-max normalized log-transformed frequency, , obtained from a representative language corpus. Min-max normalization maps a word’s log-transformed absolute frequency in the corpus, f, to a value in the range between 0 and 1, while log-transformation helps ensure a better fit for the linear model describing the relationship between a word’s rank and frequency in the corpus. For example, the WM load associated with the word ‘peace’, , after controlling for its frequency in the corpus is given by: (1) where F_max and F_min are the absolute frequencies, respectively, of the most and the least frequent word in a given corpus, and f_”peace” is the absolute frequency of the word ‘peace’ in the corpus. Estimating WM load this way seems appropriate since word frequency is a major factor determining word recognition [48]. More frequent words elicit quicker responses than less frequent ones [49,50]. In fact, Whaley [51] argued that word frequency was “by far the most powerful predictor” in a lexical decision task. For source texts we chose the Russian ruTenTen (14.5 billion words) and the British National Corpus (112.2 million words) accessed through Sketchengine interface (https://www.sketchengine.co.uk/). Before looking up a word’s frequency in the corpus, we used the NLTK package [52] to lemmatize each word.

The methods described above are predominantly based on the number of content words. However, they can be further improved by scaling each word according to its syllabic length, a potentially important factor influencing the speed of processing. This approach makes particular sense since the phonological loop—an essential component of auditory WM [53,54]—has a capacity of about 2 s unless the information stored in it is constantly rehearsed and/or processed.

In summary, we used three different methods to estimate WM load, specifically in the number of (a) content words (CW); (b) content words weighted by their frequency (CL), (c) all the words (both function and content) weighted by their respective syllabic length (SYL). In this study function words included articles, prepositions, auxiliary verbs and the conjunction “and” for English; prepositions and the conjunction “и” for Russian. Changes in WM load were assumed to occur at the time of word offset. Since the probes were delivered at random intervals not corresponding to word offsets, we used linear interpolation to estimate WM load at probe onsets (Fig 1). Given the high probing density (on average every 0.6 s), it may be assumed that the probes coincided with roughly the same ideas in the source text across participants and across directions.

Download:

Fig 1. A slice of a time-coded transcript showing WM load dynamics over time.

The blue line represents WM load at word offsets. The red line captures linearly interpolated values of WM load at probe onsets. Red numbers along the top axis show probe onset times. WM load was estimated using the CL method.

https://doi.org/10.1371/journal.pone.0206129.g001

Transcript time coding and WM load estimation

After the experiments, we made transcripts of translations produced in relation to the original source speech. These were manually time-stamped and reformatted to allow us to calculate the different WM load estimates as described above. This work was partially automated using several custom functions in VBA running under Microsoft Excel. Fig 2 illustrates the resulting data.

Download:

Fig 2. Illustration of a time-coded transcript.

WM load is estimated based on the number of content words (CW), all the words (both function and content) weighted by their min-max normalized log-transformed frequency in the respective language corpus (CL), and content words weighted by their respective syllabic length (SYL).

https://doi.org/10.1371/journal.pone.0206129.g002

EEG data acquisition and pre-processing

The electroencephalogram (EEG) was continuously recorded using the ActiCHamp recording system and BrainVision PyCorder software (Brain Products GmbH) at the sampling rate of 2000 Hz from 32 scalp sites with active electrodes seated in an elastic cap.

EEG preprocessing and analysis were carried out using the EEGlab toolbox [55] for MATLAB (The Mathworks, Natick). Raw datasets were downsampled offline to 250 Hz, converted to a linked-mastoids reference (TP9 and TP10) and bandpass-filtered from 0.1 to 30 Hz (6 db/oct) using a zero-phase FIR filter. At the next step we ran independent component analysis (ICA) using the binica algorithm in EEGlab to remove the two most dominant independent components corresponding to oculomotor artifacts. Then, to address the issue of muscle noise produced by constant articulation in the experiment, we applied the artifact subspace reconstruction (ASR) algorithm [56]. The burst parameter was set to 4 standard deviations for mild cleaning of continuous EEG data.

To ensure the possibility of averaging the data epochs within specific WM load (low, medium or high), we changed event codes in the EEG dataset to reflect WM load at the probe onset latency estimated with the three different methods described above.

The continuous EEG was then chunked into epochs of 500 ms post-stimulus (probe) onset including a 100 ms pre-stimulus baseline, yielding about 5840 epochs per subject. The epochs were screened for any residual artifacts that may have survived the ASR- and ICA-based cleaning stage. We then averaged the epochs within direction of interpretation and WM load estimated at time zero, i.e. the moment of probe onset, using each of the three methods.

Data analysis

After EEG data pre-processing, we performed several statistical tests of our hypotheses. First, we wished to check for possible main effects (and interaction) of WM load and interpretation direction on the average ERP amplitude and peak latency in the N1 and P1 range. Following the recommendations of Handy [57], we used narrow time windows surrounding the observed peaks in the grand average ERP waveform, specifically 40–80 ms (for P1) and 120–160 ms (for N1) post stimulus onset, from which window-mean amplitudes were extracted. We used a 4-way repeated measures analysis of variance (ANOVA) with Direction (L1→L2, L2→L1), WM Load (Low, Medium, High), Anteriority (Front, Center, Back) and Laterality (Left, Middle, Right) as factors. Unless stated otherwise, all the statistical tests were done in R (R Development Core Team 2016) using the ez library.

Results

WM load estimates from behavioral data

Fig 3 shows the distributions of WM loads by participant and direction of interpretation. We first consider the distributions of WM load estimated by the most straightforward CW and SYL methods described above. All the participants’ median WM loads estimated with the CW and SYL methods were smaller in the L2→L1 direction than in the opposite direction. Since all the WM load distributions were clearly right-skewed, a non-parametric test was in order. A Wilcoxon signed rank test showed this difference to be significant: V = 45, p = 0.003906. However, the difference between WM loads calculated using the other method (CL) did not reach full statistical significance: V = 6, p = 0.05469.

Download:

Fig 3. Distributions of WM load estimates by subject and direction of translation calculated using the three different methods.

https://doi.org/10.1371/journal.pone.0206129.g003

EEG results

To obtain an equal number of observations within each cell of our fully crossed experimental design, the boundaries between the high, medium and low WM load conditions should have been set at the 33^rd and 66^th quantiles of each subject’s WM load distributions (both for English-Russian and Russian-English). However, as can be inferred from Fig 3, there is a certain ‘comfort zone’ that interpreters keep to (or are forced to keep by the very conditions of the interpretation task) most of the time. Therefore, splitting the epochs into three equal WM load groups may not capture a potential effect on the neural activity due to the associated WM load. With that in mind, we labeled an epoch as low if the WM load at the probe onset was below the 10th quantile, medium if the WM load fell between the 10^th and 90^th quantile, and high if it exceeded the 90^th quantile in the corresponding WM load distribution. Although such a choice may seem arbitrary, it was a tradeoff between maximizing the effect size and sacrificing statistical power due to inflated error in the low and high WM condition. Mindful of large between-subject and between-language variance in the WM load distributions, we calculated the boundary quantiles for each participant and direction of translation individually obtaining 18 pairs (9 participants x 2 source language) of subject- and direction-specific condition boundaries.

Figs 4 and 5 show grand average ERPs at the Cz channel as a function of WM load. Fig 6 shows the topographical distributions computed on grand average ERP data. It is clearly seen that the P1 peak (~60 ms post stimulus onset) was topographically centered around frontal midline electrodes (Fz and Cz), while the N1 peak is slightly offset to the right. Accordingly, P1 and N1 amplitude and latency analyses were based on the following subset of electrodes: Fp1, Fz, F3, F7, FC5, FC1, C3, CP5, CP1, Pz, P3, P4, CP6, CP2, Cz, C4, FC6, FC2, F4, F8, Fp2.

Download:

Fig 4. Grand average ERPs for L1 (Russian) → L2 (English) and L2 (English) → L1 (Russian) direction at Cz.

The ERPs are plotted for the three different methods of estimating WM load. Y-axes display microvolts and X-axes milliseconds.

https://doi.org/10.1371/journal.pone.0206129.g004

Download:

Fig 5. Grand average P1 and N1 amplitudes at Cz as a function of WM load.

Vertical bars represent standard errors of the mean. None of the interactions reached significance. The main effect of WM load was significant for CW and CL (marked with an asterisk), see text for details.

https://doi.org/10.1371/journal.pone.0206129.g005

Download:

Fig 6. Grand average ERP topographical distributions.

The topographical distributions are shown for low (A), medium (B) and high (C) WM load. WM load estimation method: CL.

https://doi.org/10.1371/journal.pone.0206129.g006

P1 amplitude analyses.

Regardless of the WM load estimation method used, we found neither a main effect of interpretation direction (CL [F(1, 8) = 0.20, p = 0.663, η²_p = 0.02]; CW [F(1, 8) = 0.34, p = 0.577, η²_p = 0.04]; SYL [F(1, 8) = 0.52, p = 0.491, η²_p = 0.06]), nor WM Load × Direction interaction (CL [F(1.07, 8.56) = 0.19, p = 0.690, η²_p = 0.02]; CW [F(2, 16) = 0.64, p = 0.542, η²_p = 0.07]; or SYL [F(1.13, 9.03) = 0.54, p = 0.502, η²_p = 0.06]). The main effect of WM load reached significance for the two methods based on the number of content words in the source-target lag, with smaller P1 amplitudes under lower WM load (CL [F(1.23, 9.86) = 5.02, p = 0.044, η²_p = 0.39]; CW [F(2, 16) = 3.94, p = 0.041, η²_p = 0.33], but not the one based on syllables (SYL [F(2, 16) = 0.12, p = 0.887, η²_p = 0.1]). Although the effect of WM load was stronger over the frontal sites, the Anteriority × Load interaction reached significance only for the CL method [F(4, 32) = 7.64, p < 0.001, η²_p = 0.49], but not for the other methods (CW [F(1.41, 11.32) = 1.48, p = 0.261, η²_p = 0.16; SYL [F(1.57, 12.58) = 0.17, p = 0.796, η²_p = 0.02]).

N1 amplitude analyses.

Similarly to the P1 range, we found neither main effect of interpretation direction (CL [F(1, 8) = 0.03, p = 0.859, η²_p < 0.01]; CW [F(1, 8) = 0.53, p = 0.486, η²_p = 0.06]; or SYL [F(1, 8) = 1.11, p = 0.323, η²_p = 0.12]), nor WM Load × Direction interaction (CL [F(2, 16) = 0.03, p = 0.967, η²_p < 0.01]; CW [F(2,16) = 0.07, p = 0.929, η²_p < 0.01]; or SYL [F(1.18, 9.42) = 0.20, p = 0.707, η²_p = 0.02]). However, the main effect of WM load was consistently significant for all the WM load estimation methods, with larger N1 amplitudes under lower WM load (CL [F(2, 16) = 7.99, p = 0.04, η²_p = 0.50]; CW [F(2, 16) = 11.31, p < 0.001, η²_p = 0.59]; and SYL [F(2, 16) = 8.94, p = 0.002, η²_p = 0.53]). As in the P1 range, the effect of WM load was more pronounced over the frontal sites. However, the Anteriority × Load interaction did not reach significance (CL [F(1.65, 13.20) = 2.49, p = 0.127, η²_p = 0.24]; CW [F(2, 15.98) = 2.48, p = 0.115, η²_p = 0.24]; SYL [F(1.64, 13.12) = 0.94, p = 0.397, η²_p = 0.11]).

Considering the modest sample size, we performed an additional analysis of N1 and P1 amplitudes using the jackknife method as described by Miller, Ulrich and Schwartz [58]. Specifically, we took leave-one-out resamples from the data such that each of them included all but one participant’s data, and ran the same statistical tests on each resample in order to ensure that the results were not driven by a few outliers or other spurious effects; this has fully confirmed our main analysis.

P1 and N1 latency analyses.

The ANOVA on P1 peak latency data showed a significant main effect of WM load for CL, F(2, 16) = 4.90, p = 0.022, η²_p = 0.38. Specifically, P1 latency was smaller for low WM load (M = 59.66, SE = 0.043) than for high WM load (M = 60.31, SE = 0.043). No such effect was significant for CW or SYL. No other significant main effects or interactions were observed.

No significant main effects or interactions were observed from the ANOVA of N1 peak latency data for any of the WM load estimation methods.

Discussion

Although several authors have investigated functional and structural plasticity induced by SI training and practice (e.g. [59,60]), we are not aware of any attempts to validate models of SI using neuroimaging methods, and in the present study we tried to address this gap. Specifically we aimed to electrophysiologically test the Efforts Model of SI [13,15,37] which predicts that WM overload during simultaneous interpretation should decrease the amount of attention available to support the ‘listening effort’ (that the Efforts Model defines as non-automatic operations involved in auditory perception and analysis of auditory stimuli) and therefore degrade the processing of the source message. Although current WM load can be conveniently estimated if we assume it to be proportional to the size of lag between the source and the target (in the number of content words), getting a precise behavioral measure of attention during SI would have been problematic. However, as was previously demonstrated in non-interpreters, the amplitude of early ERP components time-locked to task-irrelevant probe stimuli appears more negative when the target audio is attended [17], and therefore could serve as an electrophysiological index of attention. Based on that and the prediction of the Efforts Model, we expected a larger negativity in the P1 and N1 range at small WM loads suggesting enhanced processing of the current portion of the source message.

Consistent with our hypothesis, we found that the amplitude of the P1 and N1 component elicited by the task-irrelevant pure tone probes was more negative at lower WM loads. Conversely, the P1 and N1 amplitude was less negative at higher WM loads suggesting that the brain temporarily attenuates or suspends the processing of auditory stimuli to more efficiently process and manipulate WM backlog, reduce the lag and cognitive load. The question whether this is controlled consciously (i.e. is a choice based on strategic judgment), or automatically (i.e. is a skill acquired though training and used unconsciously) has been treated in previous research (e.g. [61–63]) and still remains underexplored. A correlational study incorporating behavioral performance measures (such as WM span) on a larger sample of interpreters with different levels of experience may in the future help answer the question with more certainty.

A visual inspection of the ERP waveforms revealed that the effect of WM load in the P1/N1 range may not have been due only to enhanced negativity, but also to decreased latency of the P1 and N1 peaks for low WM load. Fehér, Folyi & Horváth [64] reported a shorter N1 latency for attended stimuli, which reinforces the case for attention acting as a latent variable mediating the effect of WM load on the early ERP components. Although our results did suggest greater P1 and N1 latencies at larger WM load, the effect was statistically significant only for the P1 range.

As part of our second research question, we wished to establish if the subjectively greater difficulty of SI from L2 into L1 reported by the respondents of a separate survey, was due to direction-specific WM load management differences and, if any, to find their electrophysiological correlates. Although the behavioral data showed a pattern of smaller median WM loads in the L2→L1 direction, which seems to be consistent with the predictions of the Efforts Model, we found no significant main effect of direction on the ERPs (P1/N1) or interaction between direction and WM load. One possible explanation could be that although on average keeping a certain number of L2 words in WM does require more effort than the same number of L1 words, the overall level of attention to the source message stays the same regardless of the language of input. In other words, in the L2→L1 direction interpreters’ attention to the source message is not inferior compared to L1→L2 direction, because the increased effort involved in processing L2 input in WM is compensated by maintaining WM load at a level lower than in L1→L2 direction.

The finding of significantly different median WM loads for the two interpretation directions may be due to several reasons. First, because L2 words on average are less ‘familiar’ than L1 words, they have weaker mental representations than the corresponding L1 words. As a result, L2 words are known to take more time to react to in lexical decision tasks [50]. For our participants (Russian L1, English L2 speakers) this indicates that the perception and analysis of L2 speech is more difficult than L1. Moreover, this difficulty is not offset by the relative ease of producing a meaningful target message in the mother tongue, which is evidenced by shorter lexical retrieval times for L1 words (e.g. [65]). Second, WM is less efficient in keeping and manipulating L2 words than L1 words [38]. Perhaps, when working from L2 into L1, interpreters conserve processing capacity by reducing WM load, while in the opposite direction they can afford to keep in WM more L1 than L2 content words. Third, interpreting from Russian into English—again, all else being equal—appears less of a challenge because Russian words are on average syllabically longer than in English [66]. Thus, a source message in Russian is informationally more sparse and easier to process than a comparable English message delivered at the same rate measured in syllables per unit time. Therefore, in English-Russian interpretation, the target Russian discourse is syllabically longer and therefore should cause more interference due to acoustic overlap with the source message. In other words, if we assume word-for-word rendering, the interpreter has less time to ‘unpack’ the English message: English words are shorter than Russian requiring more time to complete the articulation of the corresponding Russian translation. Not only does this create greater and potentially disrupting acoustic interference with the process of listening to the English source, it is possible that attention is engaged longer in the self-monitoring stage of the interpretation cycle. However, at least one study [67] based on seven languages—unfortunately, not including Russian—found that in spoken communication the average amount of information transferred per unit time is subject to substantial, although statistically insignificant, cross-language variation. It is possible that the obvious difference in WM loads as a function of source language is due to different information densities of the languages used here.

Although it is impossible to directly generalize our results to situations with any other source-target language pair, our data suggest that the relative share of attentional resources allocated to the ‘listening effort’ during SI does not depend on the language of input. We can at least argue that the main source of subjective difficulty in the English-Russian interpreting does not lie in input language-specific distribution of attention.

While enhanced negativity in the P1/N1 range under low WM load during SI fits well with the Efforts Model and task-engagement/distraction trade-off model [68,69], there is room for another interpretation. It is possible that when WM load increases, the task-irrelevant probes are processed less, while the task-relevant source message still receives the same amount of attention as it does under low WM load. This could be an interesting subject of a follow-up investigation.

Although the number of participants in the present study was relatively small, by limiting our selection to professionally trained highly-skilled individuals for whom interpreting is a main occupation rather than a side activity, we believe that we could focus on a population that can in fact be characterized as a core group of conference interpreters, rather than mere L2 speakers tasked with an interpreting job. Notably, we were able to verify the results using the more elaborate jacknife analytical statistics tools which fully confirmed our original analysis; that said, future studies will benefit from attempting to collect larger samples of professional interpreters, albeit this is not a trivial task.

On a practical note, our results justify a recommendation to keep WM load within reasonable limits during SI. The question about how this can be achieved is very relevant. Several studies [7,70,71] have shown that simultaneous interpreters use a range of strategies to manage their processing load (e.g. omitting redundancies in the source speech). None of them, however, have attempted to identify the neural states that determine—or at least bias—the choice of a particular strategy.

Further neuroimaging studies are needed to validate and/or refine theoretical models of SI and will help to guide interpreters, SI instructors and the interpretation industry towards better interpretation practices, more efficient curricula and higher standards of service.

Limitations and future directions

One major limitation of the study is the narrow group of participants that only included L1 Russian/L2 English participants. While our results do lend support to the Efforts Model of simultaneous interpreting, one should be cautious in directly generalizing its results to any other subpopulations of conference interpreters whose L1 and L2 are other than Russian or English. Every language presents a unique set of difficulties at multiple levels, including syntactic, lexical and phonetic, which potentially influence interpreting strategies including the lag and overall working memory load. For example, when working from German, interpreters are forced to increase their décalage [12] waiting for the predicate that often appears at the very end of the clause, which means a larger WM load overall. An especially interesting direction for future research is to conduct similar experiment on L1 English speakers translating the same speeches to see if the pattern of results would be consistent with the predictions of the Efforts Model.

Another limitation of the study relates to possible performance differences that may exist across the different interpreters. To address this limitation, future research could implement a more advanced design incorporating behavioral performance measures (e.g. working memory span and speed of lexical access) into the overall data analysis and interpretation of results. It would be interesting to see if these performance measures are correlated with average WM load during the performance of an SI task.

Finally, it has to be acknowledged that our study did not analyze the role of the speaking effort–the third essential component of the Efforts Model–not least because of the severe difficulties of overcoming articulation artifacts in EEG analysis. Future EEG studies may provide more valuable insights into the microdynamics of cognitive processes during simultaneous interpreting and help advance the state of the art in conference interpreter training.

Supporting information

S1 Appendix. Parameters of speeches used in the interpreting task.

https://doi.org/10.1371/journal.pone.0206129.s001

(DOCX)

S2 Appendix. The order in which the speeches were presented.

https://doi.org/10.1371/journal.pone.0206129.s002

(DOCX)

References

1. Proverbio AM, Leoni G, Zani A. Language switching mechanisms in simultaneous interpreters: an ERP study. Neuropsychologia. 2004;42: 1636–1656. pmid:15327931
- View Article
- PubMed/NCBI
- Google Scholar
2. Bajo MT, Padilla F, Padilla P. Comprehension processes in simultaneous interpreting [Internet]. Chesterman A, Salvador NGS, Gambier Y, editors. Translation in Context: Selected papers from the EST Congress, Granada 1998. John Benjamins Publishing Company; 2000. https://doi.org/10.1075/btl.39.15baj
3. Lambert S. Shadowing. Meta J des traducteurs. 1992;37: 263.
- View Article
- Google Scholar
4. Padilla P, Bajo MT, Cañas JJ, Padilla F. Cognitive processes of memory in simultaneous interpretation. Topics in interpreting research. 1995. pp. 61–71.
- View Article
- Google Scholar
5. Chernov G V. Inference and Anticipation in Simultaneous Interpreting [Internet]. Setton R, Hild A, editors. Amsterdam: John Benjamins Publishing Company; 2004. https://doi.org/10.1075/btl.57
6. Kim H-R. Linguistic Characteristics and Interpretation Strategy Based on EVS Analysis of Korean-Chinese, Korean-Japanese Interpretation. Meta Transl J. 2005;50.
- View Article
- Google Scholar
7. Li C. Coping Strategies for Fast Delivery in Simultaneous Interpretation. J Spec Transl. 2010; 19–25.
- View Article
- Google Scholar
8. Liu M, Schallert DL, Carroll PJ. Working memory and expertise in simultaneous interpreting. Interpreting. 2004;6: 19–42.
- View Article
- Google Scholar
9. Mizuno A. Process Model for Simultaneous Interpreting and Working Memory. Meta J des traducteurs. 2005;50: 739.
- View Article
- Google Scholar
10. Pym A. On omission in simultaneous interpreting. Risk analysis of a hidden effort. Efforts Model Interpret Transl Res. 2008; 83–105.
- View Article
- Google Scholar
11. Signorelli TM, Obler L. Working memory in simultaneous interpreters. Memory, language, and bilingualism: Theoretical and applied approaches. 2013. pp. 95–125. Available: http://search.ebscohost.com/login.aspx?direct=true&db=psyh&AN=2012-26865-004&site=ehost-live&scope=site
- View Article
- Google Scholar
12. Goldman-Eisler F. Segmentation of input in simultaneous translation. J Psycholinguist Res. 1972;1: 127–40. pmid:24197574
- View Article
- PubMed/NCBI
- Google Scholar
13. Gile D. Le partage de l’attention et le “modèle d’effort” en interprétation simultaneé. Interpret Newsl. Università degli Studi di Trieste-Scuola Superiore di Lingue Moderne per Intrepreti e Traduttor; 1988; 4–22. Available: http://hdl.handle.net/10077/2132
14. Escera C, Alho K, Winkler I, Näätänen R. Neural Mechanisms of Involuntary Attention to Acoustic Novelty and Change. J Cogn Neurosci. 1998;10: 590–604. pmid:9802992
- View Article
- PubMed/NCBI
- Google Scholar
15. Gile D. Testing the Effort Models’ tightrope hypothesis in simultaneous interpreting-A contribution. Hermes (Wiesb). 1999; 153–172.
- View Article
- Google Scholar
16. Giuliano RJ, Karns CM, Neville HJ, Hillyard SA. Early Auditory Evoked Potential Is Modulated by Selective Attention and Related to Individual Differences in Visual Working Memory Capacity. J Cogn Neurosci. 2014;26: 2682–2690. pmid:25000526
- View Article
- PubMed/NCBI
- Google Scholar
17. Näätanen R. Processing negativity: An evoked-potential reflection. Psychol Bull. 1982;92: 605–640. pmid:7156260
- View Article
- PubMed/NCBI
- Google Scholar
18. Coch D, Sanders LD, Neville HJ. An event-related potential study of selective auditory attention in children and adults. J Cogn Neurosci. 2005;17: 605–622. pmid:15829081
- View Article
- PubMed/NCBI
- Google Scholar
19. Hillyard S a, Hink RF, Schwent VL, Picton TW. Electrical Signs of Selective Attention in the Human Brain. Science (80-). 1973;182: 177–180.
- View Article
- Google Scholar
20. Hink RF, Hillyard S a. Auditory evoked potentials during selective listening to dichotic speech messages. Percept Psychophys. 1976;20: 236–242.
- View Article
- Google Scholar
21. Teder W, Alho K, Reinikainen K, Näätänen R. Interstimulus interval and the selective-attention effect on auditory ERPs: Psychophysiology. 1993;30: 71–81. pmid:8416064
- View Article
- PubMed/NCBI
- Google Scholar
22. Woldorff MG, Hillyard S a. Modulation of early auditory processing during selective listening to rapidly presented tones. Electroencephalogr Clin Neurophysiol. 1991;79: 170–191. pmid:1714809
- View Article
- PubMed/NCBI
- Google Scholar
23. Woods DL, Hillyard S a, Hansen JC. Event-related brain potentials reveal similar attentional mechanisms during selective listening and shadowing. J Exp Psychol Hum Percept Perform. 1984;10: 761–77. pmid:6239905
- View Article
- PubMed/NCBI
- Google Scholar
24. Pratt N, Willoughby A, Swick D. Effects of Working Memory Load on Visual Selective Attention: Behavioral and Electrophysiological Evidence. Front Hum Neurosci. 2011;5: 57. pmid:21716633
- View Article
- PubMed/NCBI
- Google Scholar
25. Camayd-Freixas E. Cognitive theory of simultaneous interpreting and training. Proc 52nd Conf Am Transl Assoc. 2011; 13. Available: http://casgroup.fiu.edu/pages/docs/2406/1320876652_Cognitive_Theory_of_Simultation_Interpreting.pdf
- View Article
- Google Scholar
26. Sabri M, Humphries C, Verber M, Liebenthal E, Binder JR, Mangalathu J, et al. Neural effects of cognitive control load on auditory selective attention. Neuropsychologia. 2014;61: 269–279. pmid:24946314
- View Article
- PubMed/NCBI
- Google Scholar
27. Awh E, Vogel EK, Oh S-H. Interactions between attention and working memory. Neuroscience. 2006;139: 201–208. pmid:16324792
- View Article
- PubMed/NCBI
- Google Scholar
28. Awh E, Jonides J. Overlapping mechanisms of attention and spatial working memory. Trends Cogn Sci. 2001;5: 119–126. pmid:11239812
- View Article
- PubMed/NCBI
- Google Scholar
29. Corbetta M, Shulman GL. Control of Goal-Directed and Stimulus-Driven Attention in the Brain. Nat Rev Neurosci. 2002;3: 215–229. pmid:11994752
- View Article
- PubMed/NCBI
- Google Scholar
30. Cowan N. Processing limits of selective attention and working memory: Potential implications for interpreting. Interpreting. 2000;5: 117–146.
- View Article
- Google Scholar
31. Mayer JS, Bittner R a., Nikolić D, Bledowski C, Goebel R, Linden DEJ. Common neural substrates for visual working memory and attention. Neuroimage. 2007;36: 441–453. pmid:17462914
- View Article
- PubMed/NCBI
- Google Scholar
32. Ungerleider SK and LG. Mechanisms of Visual Attention in the Human Cortex. Annu Rev Neurosci. 2000;23: 315–341. pmid:10845067
- View Article
- PubMed/NCBI
- Google Scholar
33. Zanto TP, Rubens MT, Thangavel A, Gazzaley A. Causal role of the prefrontal cortex in top-down modulation of visual processing and working memory. Nat Neurosci. Nature Publishing Group; 2011;14: 656–661. pmid:21441920
- View Article
- PubMed/NCBI
- Google Scholar
34. NICODEMUS B, EMMOREY K. Direction asymmetries in spoken and signed language interpreting. Biling Lang Cogn. 2013;16: 624–636. pmid:23833563
- View Article
- PubMed/NCBI
- Google Scholar
35. Rinne JO, Tommola J, Laine M, Krause BJ, Schmidt D, Kaasinen V, et al. The translating brain: cerebral activation patterns during simultaneous interpreting. Neurosci Lett. 2000;294: 85–8. pmid:11058793
- View Article
- PubMed/NCBI
- Google Scholar
36. Al-Salman S, Al-Khanji R. The Native Language Factor in Simultaneous Interpretation in an Arabic/English Context. Meta J des traducteurs. 2002;47: 607.
- View Article
- Google Scholar
37. Gile D. Basic Concepts and Models for Interpreter and Translator Training [Internet]. Amsterdam: John Benjamins Publishing Company; 1995. https://doi.org/10.1075/btl.8(1st)
38. Ardila A. Language representation and working memory with bilinguals. J Commun Disord. 2003;36: 233–240. pmid:12742670
- View Article
- PubMed/NCBI
- Google Scholar
39. Thorn ASC, Gathercole SE. Language differences in verbal short-term memory do not exclusively originate in the process of subvocal rehearsal. Psychon Bull Rev. 2001;8: 357–364. pmid:11495126
- View Article
- PubMed/NCBI
- Google Scholar
40. Gardner MK, Rothkopf EZ, Lapan R, Lafferty T. The word frequency effect in lexical decision: Finding a frequency-based component. Mem Cognit. 1987;15: 24–28. pmid:3821488
- View Article
- PubMed/NCBI
- Google Scholar
41. Rubenstein H, Lewis SS, Rubenstein MA. Evidence for phonemic recoding in visual word recognition. J Verbal Learning Verbal Behav. 1971;10: 645–657.
- View Article
- Google Scholar
42. Scarborough D, Cortese C, Scarborough HS. Frequency and repetition effects in lexical memory. J Exp Psychol Gen. 1977;3: 1–17.
- View Article
- Google Scholar
43. Gerver D. The Effects of Noise on the Performance of Simultaneous Interpreters: Accuracy of Performance. Acta Psychol (Amst). 1974; 159–167.
- View Article
- Google Scholar
44. Gerver D. A psychological approach to simultaneous interpretation. Meta. 1975;20/22: 119–128.
- View Article
- Google Scholar
45. Gile D. Local Cognitive Load in Simultaneous Interpreting And Its Implications For Empirical Research. Forum Fam Plan West Hemisph. 2008;6: 59–77.
- View Article
- Google Scholar
46. Setton R. Simultaneous interpretation: a cognitive-pragmatic analysis. Benjamins translation library. 1999.
- View Article
- Google Scholar
47. Peirce JW. Generating Stimuli for Neuroscience Using PsychoPy. Front Neuroinform. 2008;2: 10. pmid:19198666
- View Article
- PubMed/NCBI
- Google Scholar
48. Burgess C, Livesay K. The effect of corpus size in predicting reaction time in a basic word recognition task: Moving on from Kučera and Francis. Behav Res Methods, Instruments, Comput. 1998;30: 272–277.
- View Article
- Google Scholar
49. Dobbs AR, Friedman A, Lloyd J. Frequency effects in lexical decisions: A test of the verification model. J Exp Psychol Hum Percept Perform. 1985;11: 81–92.
- View Article
- Google Scholar
50. Gernsbacher MA. Resolving 20 Years of Inconsistent Interactions Between Lexical Familiarity and Orthography, Concreteness, and Polysemy. J Exp Psychol Gen. 1984;113: 256–281. pmid:6242753
- View Article
- PubMed/NCBI
- Google Scholar
51. Whaley CP. Word—nonword classification time. J Verbal Learning Verbal Behav. 1978;17: 143–154.
- View Article
- Google Scholar
52. Bird S, Klein E, Loper E. Natural Language Processing with Python [Internet]. O’Reilly Media; 2009. Available: http://books.google.com/books?id=KGIbfiiP1i4C&printsec=frontcover
- View Article
- Google Scholar
53. Baddeley A. Working memory. Science (80-). 1992; Available: http://www.sciencemag.org/content/255/5044/556.short
- View Article
- Google Scholar
54. Baddeley A, Hitch G. Working memory. Bower GH, editor. Psychol Learn Motiv. New York: Academic Press; 1974;8: 47–89. Available: https://www.google.com/books?hl=en&lr=&id=o5LScJ9ecGUC&oi=fnd&pg=PA47&dq=Working+Memory&ots=8z8G_U5gZ_&sig=c3N0a_Kl0hExGJW8aPgZIcpNG84
55. Delorme A, Makeig S. EEGLAB: An open source toolbox for analysis of single-trial EEG dynamics including independent component analysis. J Neurosci Methods. 2004;134: 9–21. pmid:15102499
- View Article
- PubMed/NCBI
- Google Scholar
56. Mullen TR, Kothe CAE, Chi YM, Ojeda A, Kerth T, Makeig S, et al. Real-time neuroimaging and cognitive monitoring using wearable dry EEG. IEEE Trans Biomed Eng. 2015;62: 2553–2567. pmid:26415149
- View Article
- PubMed/NCBI
- Google Scholar
57. Handy TC. Basic Principles of ERP Quantification. In: Handy TC, editor. Event-related Potentials: A Methods Handbook. Cambrige: MIT Press; 2005.
58. Miller J, Ulrich R, Schwarz W. Why jackknifing yields good latency estimates. Psychophysiology. 2009;46: 300–312. pmid:19207197
- View Article
- PubMed/NCBI
- Google Scholar
59. Elmer S, Hänggi J, Jäncke L. Processing demands upon cognitive, linguistic, and articulatory functions promote grey matter plasticity in the adult multilingual brain: Insights from simultaneous interpreters. Cortex. 2014;54: 179–189. pmid:24699036
- View Article
- PubMed/NCBI
- Google Scholar
60. Elmer S, Kühnis J. Functional Connectivity in the Left Dorsal Stream Facilitates Simultaneous Language Translation: An EEG Study. Front Hum Neurosci. 2016;10: 1–9.
- View Article
- Google Scholar
61. Oleron P, Nanpon H. Recherches sur la traduction simultanée. Journal de Psychologie normale et pathologique. 1965. pp. 73–94. pmid:14302791
- View Article
- PubMed/NCBI
- Google Scholar
62. Kade O, Cartellieri C. Some Methodological Aspects of Simultaneous Interpreting. Babel. 1971;17: 12–16.
- View Article
- Google Scholar
63. Chincotta D, Underwood G. Simultaneous interpreters and the effect of concurrent articulation on immediate memory: A bilingual digit span study. Interpreting. 1998;3: 1–20.
- View Article
- Google Scholar
64. Fehér B, Folyi T, Horváth J. Attention decreases auditory N1 latency. Front Neurosci Conference Abstract: 13th Conference of the Hungarian Neuroscience Society. 2011.
- View Article
- Google Scholar
65. Schoonen R, Gelderen A Van, Glopper K de, Hulstijn J, Simis A, Snellings P, et al. First Language and Second Language Writing: The Role of Linguistic Knowledge, Speed of Processing, and Metacognitive Knowledge. Lang Learn. 2003;53: 165–202.
- View Article
- Google Scholar
66. Gurin I. Priemy rechevoi kompressii pri sinkhronnom perevode c russikogo yazyka an angliiskii [Speech Compression Techniques For Simultaneous Interpetation From Russian Into English]. (PhD Thesis). Moscow State University; 2009.
67. Pellegrino F, Coupé C, Marsico E. Across-Language Perspective on Speech Information Rate. Language (Baltim). 2011;87: 539–558.
- View Article
- Google Scholar
68. Simon SS, Tusch ES, Holcomb PJ, Daffner KR. Increasing Working Memory Load Reduces Processing of Cross-Modal Task-Irrelevant Stimuli Even after Controlling for Task Difficulty and Executive Capacity. Front Hum Neurosci. 2016;10: 1–13.
- View Article
- Google Scholar
69. Sörqvist P, Rönnberg J. Individual differences in distractibility: An update and a model. PsyCh J. 2014;3: 42–57. pmid:25632345
- View Article
- PubMed/NCBI
- Google Scholar
70. Bartłomiejczyk M. Strategies of simultaneous interpreting and directionality. Interpreting. 2006;8: 149–174.
- View Article
- Google Scholar
71. Chernov G V. Message redundancy and message anticipation in simultaneous interpreting. In: Lambert S, Moser-Mercer B, editors. Bridging the Gap: Empirical research in simultaneous interpretation. Chapel Hill: John Benjamins Publishing Co.; 1994. pp. 139–153. https://doi.org/10.1075/btl.3.13che

[ref1] 1. Proverbio AM, Leoni G, Zani A. Language switching mechanisms in simultaneous interpreters: an ERP study. Neuropsychologia. 2004;42: 1636–1656. pmid:15327931
View Article
PubMed/NCBI
Google Scholar

[2] View Article

[3] PubMed/NCBI

[4] Google Scholar

[ref2] 2. Bajo MT, Padilla F, Padilla P. Comprehension processes in simultaneous interpreting [Internet]. Chesterman A, Salvador NGS, Gambier Y, editors. Translation in Context: Selected papers from the EST Congress, Granada 1998. John Benjamins Publishing Company; 2000. https://doi.org/10.1075/btl.39.15baj

[ref3] 3. Lambert S. Shadowing. Meta J des traducteurs. 1992;37: 263.
View Article
Google Scholar

[7] View Article

[8] Google Scholar

[ref4] 4. Padilla P, Bajo MT, Cañas JJ, Padilla F. Cognitive processes of memory in simultaneous interpretation. Topics in interpreting research. 1995. pp. 61–71.
View Article
Google Scholar

[10] View Article

[11] Google Scholar

[ref5] 5. Chernov G V. Inference and Anticipation in Simultaneous Interpreting [Internet]. Setton R, Hild A, editors. Amsterdam: John Benjamins Publishing Company; 2004. https://doi.org/10.1075/btl.57

[ref6] 6. Kim H-R. Linguistic Characteristics and Interpretation Strategy Based on EVS Analysis of Korean-Chinese, Korean-Japanese Interpretation. Meta Transl J. 2005;50.
View Article
Google Scholar

[14] View Article

[15] Google Scholar

[ref7] 7. Li C. Coping Strategies for Fast Delivery in Simultaneous Interpretation. J Spec Transl. 2010; 19–25.
View Article
Google Scholar

[17] View Article

[18] Google Scholar

[ref8] 8. Liu M, Schallert DL, Carroll PJ. Working memory and expertise in simultaneous interpreting. Interpreting. 2004;6: 19–42.
View Article
Google Scholar

[20] View Article

[21] Google Scholar

[ref9] 9. Mizuno A. Process Model for Simultaneous Interpreting and Working Memory. Meta J des traducteurs. 2005;50: 739.
View Article
Google Scholar

[23] View Article

[24] Google Scholar

[ref10] 10. Pym A. On omission in simultaneous interpreting. Risk analysis of a hidden effort. Efforts Model Interpret Transl Res. 2008; 83–105.
View Article
Google Scholar

[26] View Article

[27] Google Scholar

[ref11] 11. Signorelli TM, Obler L. Working memory in simultaneous interpreters. Memory, language, and bilingualism: Theoretical and applied approaches. 2013. pp. 95–125. Available: http://search.ebscohost.com/login.aspx?direct=true&db=psyh&AN=2012-26865-004&site=ehost-live&scope=site
View Article
Google Scholar

[29] View Article

[30] Google Scholar

[ref12] 12. Goldman-Eisler F. Segmentation of input in simultaneous translation. J Psycholinguist Res. 1972;1: 127–40. pmid:24197574
View Article
PubMed/NCBI
Google Scholar

[32] View Article

[33] PubMed/NCBI

[34] Google Scholar

[ref13] 13. Gile D. Le partage de l’attention et le “modèle d’effort” en interprétation simultaneé. Interpret Newsl. Università degli Studi di Trieste-Scuola Superiore di Lingue Moderne per Intrepreti e Traduttor; 1988; 4–22. Available: http://hdl.handle.net/10077/2132

[ref14] 14. Escera C, Alho K, Winkler I, Näätänen R. Neural Mechanisms of Involuntary Attention to Acoustic Novelty and Change. J Cogn Neurosci. 1998;10: 590–604. pmid:9802992
View Article
PubMed/NCBI
Google Scholar

[37] View Article

[38] PubMed/NCBI

[39] Google Scholar

[ref15] 15. Gile D. Testing the Effort Models’ tightrope hypothesis in simultaneous interpreting-A contribution. Hermes (Wiesb). 1999; 153–172.
View Article
Google Scholar

[41] View Article

[42] Google Scholar

[ref16] 16. Giuliano RJ, Karns CM, Neville HJ, Hillyard SA. Early Auditory Evoked Potential Is Modulated by Selective Attention and Related to Individual Differences in Visual Working Memory Capacity. J Cogn Neurosci. 2014;26: 2682–2690. pmid:25000526
View Article
PubMed/NCBI
Google Scholar

[44] View Article

[45] PubMed/NCBI

[46] Google Scholar

[ref17] 17. Näätanen R. Processing negativity: An evoked-potential reflection. Psychol Bull. 1982;92: 605–640. pmid:7156260
View Article
PubMed/NCBI
Google Scholar

[48] View Article

[49] PubMed/NCBI

[50] Google Scholar

[ref18] 18. Coch D, Sanders LD, Neville HJ. An event-related potential study of selective auditory attention in children and adults. J Cogn Neurosci. 2005;17: 605–622. pmid:15829081
View Article
PubMed/NCBI
Google Scholar

[52] View Article

[53] PubMed/NCBI

[54] Google Scholar

[ref19] 19. Hillyard S a, Hink RF, Schwent VL, Picton TW. Electrical Signs of Selective Attention in the Human Brain. Science (80-). 1973;182: 177–180.
View Article
Google Scholar

[56] View Article

[57] Google Scholar

[ref20] 20. Hink RF, Hillyard S a. Auditory evoked potentials during selective listening to dichotic speech messages. Percept Psychophys. 1976;20: 236–242.
View Article
Google Scholar

[59] View Article

[60] Google Scholar

[ref21] 21. Teder W, Alho K, Reinikainen K, Näätänen R. Interstimulus interval and the selective-attention effect on auditory ERPs: Psychophysiology. 1993;30: 71–81. pmid:8416064
View Article
PubMed/NCBI
Google Scholar

[62] View Article

[63] PubMed/NCBI

[64] Google Scholar

[ref22] 22. Woldorff MG, Hillyard S a. Modulation of early auditory processing during selective listening to rapidly presented tones. Electroencephalogr Clin Neurophysiol. 1991;79: 170–191. pmid:1714809
View Article
PubMed/NCBI
Google Scholar

[66] View Article

[67] PubMed/NCBI

[68] Google Scholar

[ref23] 23. Woods DL, Hillyard S a, Hansen JC. Event-related brain potentials reveal similar attentional mechanisms during selective listening and shadowing. J Exp Psychol Hum Percept Perform. 1984;10: 761–77. pmid:6239905
View Article
PubMed/NCBI
Google Scholar

[70] View Article

[71] PubMed/NCBI

[72] Google Scholar

[ref24] 24. Pratt N, Willoughby A, Swick D. Effects of Working Memory Load on Visual Selective Attention: Behavioral and Electrophysiological Evidence. Front Hum Neurosci. 2011;5: 57. pmid:21716633
View Article
PubMed/NCBI
Google Scholar

[74] View Article

[75] PubMed/NCBI

[76] Google Scholar

[ref25] 25. Camayd-Freixas E. Cognitive theory of simultaneous interpreting and training. Proc 52nd Conf Am Transl Assoc. 2011; 13. Available: http://casgroup.fiu.edu/pages/docs/2406/1320876652_Cognitive_Theory_of_Simultation_Interpreting.pdf
View Article
Google Scholar

[78] View Article

[79] Google Scholar

[ref26] 26. Sabri M, Humphries C, Verber M, Liebenthal E, Binder JR, Mangalathu J, et al. Neural effects of cognitive control load on auditory selective attention. Neuropsychologia. 2014;61: 269–279. pmid:24946314
View Article
PubMed/NCBI
Google Scholar

[81] View Article

[82] PubMed/NCBI

[83] Google Scholar

[ref27] 27. Awh E, Vogel EK, Oh S-H. Interactions between attention and working memory. Neuroscience. 2006;139: 201–208. pmid:16324792
View Article
PubMed/NCBI
Google Scholar

[85] View Article

[86] PubMed/NCBI

[87] Google Scholar

[ref28] 28. Awh E, Jonides J. Overlapping mechanisms of attention and spatial working memory. Trends Cogn Sci. 2001;5: 119–126. pmid:11239812
View Article
PubMed/NCBI
Google Scholar

[89] View Article

[90] PubMed/NCBI

[91] Google Scholar

[ref29] 29. Corbetta M, Shulman GL. Control of Goal-Directed and Stimulus-Driven Attention in the Brain. Nat Rev Neurosci. 2002;3: 215–229. pmid:11994752
View Article
PubMed/NCBI
Google Scholar

[93] View Article

[94] PubMed/NCBI

[95] Google Scholar

[ref30] 30. Cowan N. Processing limits of selective attention and working memory: Potential implications for interpreting. Interpreting. 2000;5: 117–146.
View Article
Google Scholar

[97] View Article

[98] Google Scholar

[ref31] 31. Mayer JS, Bittner R a., Nikolić D, Bledowski C, Goebel R, Linden DEJ. Common neural substrates for visual working memory and attention. Neuroimage. 2007;36: 441–453. pmid:17462914
View Article
PubMed/NCBI
Google Scholar

[100] View Article

[101] PubMed/NCBI

[102] Google Scholar

[ref32] 32. Ungerleider SK and LG. Mechanisms of Visual Attention in the Human Cortex. Annu Rev Neurosci. 2000;23: 315–341. pmid:10845067
View Article
PubMed/NCBI
Google Scholar

[104] View Article

[105] PubMed/NCBI

[106] Google Scholar

[ref33] 33. Zanto TP, Rubens MT, Thangavel A, Gazzaley A. Causal role of the prefrontal cortex in top-down modulation of visual processing and working memory. Nat Neurosci. Nature Publishing Group; 2011;14: 656–661. pmid:21441920
View Article
PubMed/NCBI
Google Scholar

[108] View Article

[109] PubMed/NCBI

[110] Google Scholar

[ref34] 34. NICODEMUS B, EMMOREY K. Direction asymmetries in spoken and signed language interpreting. Biling Lang Cogn. 2013;16: 624–636. pmid:23833563
View Article
PubMed/NCBI
Google Scholar

[112] View Article

[113] PubMed/NCBI

[114] Google Scholar

[ref35] 35. Rinne JO, Tommola J, Laine M, Krause BJ, Schmidt D, Kaasinen V, et al. The translating brain: cerebral activation patterns during simultaneous interpreting. Neurosci Lett. 2000;294: 85–8. pmid:11058793
View Article
PubMed/NCBI
Google Scholar

[116] View Article

[117] PubMed/NCBI

[118] Google Scholar

[ref36] 36. Al-Salman S, Al-Khanji R. The Native Language Factor in Simultaneous Interpretation in an Arabic/English Context. Meta J des traducteurs. 2002;47: 607.
View Article
Google Scholar

[120] View Article

[121] Google Scholar

[ref37] 37. Gile D. Basic Concepts and Models for Interpreter and Translator Training [Internet]. Amsterdam: John Benjamins Publishing Company; 1995. https://doi.org/10.1075/btl.8(1st)

[ref38] 38. Ardila A. Language representation and working memory with bilinguals. J Commun Disord. 2003;36: 233–240. pmid:12742670
View Article
PubMed/NCBI
Google Scholar

[124] View Article

[125] PubMed/NCBI

[126] Google Scholar

[ref39] 39. Thorn ASC, Gathercole SE. Language differences in verbal short-term memory do not exclusively originate in the process of subvocal rehearsal. Psychon Bull Rev. 2001;8: 357–364. pmid:11495126
View Article
PubMed/NCBI
Google Scholar

[128] View Article

[129] PubMed/NCBI

[130] Google Scholar

[ref40] 40. Gardner MK, Rothkopf EZ, Lapan R, Lafferty T. The word frequency effect in lexical decision: Finding a frequency-based component. Mem Cognit. 1987;15: 24–28. pmid:3821488
View Article
PubMed/NCBI
Google Scholar

[132] View Article

[133] PubMed/NCBI

[134] Google Scholar

[ref41] 41. Rubenstein H, Lewis SS, Rubenstein MA. Evidence for phonemic recoding in visual word recognition. J Verbal Learning Verbal Behav. 1971;10: 645–657.
View Article
Google Scholar

[136] View Article

[137] Google Scholar

[ref42] 42. Scarborough D, Cortese C, Scarborough HS. Frequency and repetition effects in lexical memory. J Exp Psychol Gen. 1977;3: 1–17.
View Article
Google Scholar

[139] View Article

[140] Google Scholar

[ref43] 43. Gerver D. The Effects of Noise on the Performance of Simultaneous Interpreters: Accuracy of Performance. Acta Psychol (Amst). 1974; 159–167.
View Article
Google Scholar

[142] View Article

[143] Google Scholar

[ref44] 44. Gerver D. A psychological approach to simultaneous interpretation. Meta. 1975;20/22: 119–128.
View Article
Google Scholar

[145] View Article

[146] Google Scholar

[ref45] 45. Gile D. Local Cognitive Load in Simultaneous Interpreting And Its Implications For Empirical Research. Forum Fam Plan West Hemisph. 2008;6: 59–77.
View Article
Google Scholar

[148] View Article

[149] Google Scholar

[ref46] 46. Setton R. Simultaneous interpretation: a cognitive-pragmatic analysis. Benjamins translation library. 1999.
View Article
Google Scholar

[151] View Article

[152] Google Scholar

[ref47] 47. Peirce JW. Generating Stimuli for Neuroscience Using PsychoPy. Front Neuroinform. 2008;2: 10. pmid:19198666
View Article
PubMed/NCBI
Google Scholar

[154] View Article

[155] PubMed/NCBI

[156] Google Scholar

[ref48] 48. Burgess C, Livesay K. The effect of corpus size in predicting reaction time in a basic word recognition task: Moving on from Kučera and Francis. Behav Res Methods, Instruments, Comput. 1998;30: 272–277.
View Article
Google Scholar

[158] View Article

[159] Google Scholar

[ref49] 49. Dobbs AR, Friedman A, Lloyd J. Frequency effects in lexical decisions: A test of the verification model. J Exp Psychol Hum Percept Perform. 1985;11: 81–92.
View Article
Google Scholar

[161] View Article

[162] Google Scholar

[ref50] 50. Gernsbacher MA. Resolving 20 Years of Inconsistent Interactions Between Lexical Familiarity and Orthography, Concreteness, and Polysemy. J Exp Psychol Gen. 1984;113: 256–281. pmid:6242753
View Article
PubMed/NCBI
Google Scholar

[164] View Article

[165] PubMed/NCBI

[166] Google Scholar

[ref51] 51. Whaley CP. Word—nonword classification time. J Verbal Learning Verbal Behav. 1978;17: 143–154.
View Article
Google Scholar

[168] View Article

[169] Google Scholar

[ref52] 52. Bird S, Klein E, Loper E. Natural Language Processing with Python [Internet]. O’Reilly Media; 2009. Available: http://books.google.com/books?id=KGIbfiiP1i4C&printsec=frontcover
View Article
Google Scholar

[171] View Article

[172] Google Scholar

[ref53] 53. Baddeley A. Working memory. Science (80-). 1992; Available: http://www.sciencemag.org/content/255/5044/556.short
View Article
Google Scholar

[174] View Article

[175] Google Scholar

[ref54] 54. Baddeley A, Hitch G. Working memory. Bower GH, editor. Psychol Learn Motiv. New York: Academic Press; 1974;8: 47–89. Available: https://www.google.com/books?hl=en&lr=&id=o5LScJ9ecGUC&oi=fnd&pg=PA47&dq=Working+Memory&ots=8z8G_U5gZ_&sig=c3N0a_Kl0hExGJW8aPgZIcpNG84

[ref55] 55. Delorme A, Makeig S. EEGLAB: An open source toolbox for analysis of single-trial EEG dynamics including independent component analysis. J Neurosci Methods. 2004;134: 9–21. pmid:15102499
View Article
PubMed/NCBI
Google Scholar

[178] View Article

[179] PubMed/NCBI

[180] Google Scholar

[ref56] 56. Mullen TR, Kothe CAE, Chi YM, Ojeda A, Kerth T, Makeig S, et al. Real-time neuroimaging and cognitive monitoring using wearable dry EEG. IEEE Trans Biomed Eng. 2015;62: 2553–2567. pmid:26415149
View Article
PubMed/NCBI
Google Scholar

[182] View Article

[183] PubMed/NCBI

[184] Google Scholar

[ref57] 57. Handy TC. Basic Principles of ERP Quantification. In: Handy TC, editor. Event-related Potentials: A Methods Handbook. Cambrige: MIT Press; 2005.

[ref58] 58. Miller J, Ulrich R, Schwarz W. Why jackknifing yields good latency estimates. Psychophysiology. 2009;46: 300–312. pmid:19207197
View Article
PubMed/NCBI
Google Scholar

[187] View Article

[188] PubMed/NCBI

[189] Google Scholar

[ref59] 59. Elmer S, Hänggi J, Jäncke L. Processing demands upon cognitive, linguistic, and articulatory functions promote grey matter plasticity in the adult multilingual brain: Insights from simultaneous interpreters. Cortex. 2014;54: 179–189. pmid:24699036
View Article
PubMed/NCBI
Google Scholar

[191] View Article

[192] PubMed/NCBI

[193] Google Scholar

[ref60] 60. Elmer S, Kühnis J. Functional Connectivity in the Left Dorsal Stream Facilitates Simultaneous Language Translation: An EEG Study. Front Hum Neurosci. 2016;10: 1–9.
View Article
Google Scholar

[195] View Article

[196] Google Scholar

[ref61] 61. Oleron P, Nanpon H. Recherches sur la traduction simultanée. Journal de Psychologie normale et pathologique. 1965. pp. 73–94. pmid:14302791
View Article
PubMed/NCBI
Google Scholar

[198] View Article

[199] PubMed/NCBI

[200] Google Scholar

[ref62] 62. Kade O, Cartellieri C. Some Methodological Aspects of Simultaneous Interpreting. Babel. 1971;17: 12–16.
View Article
Google Scholar

[202] View Article

[203] Google Scholar

[ref63] 63. Chincotta D, Underwood G. Simultaneous interpreters and the effect of concurrent articulation on immediate memory: A bilingual digit span study. Interpreting. 1998;3: 1–20.
View Article
Google Scholar

[205] View Article

[206] Google Scholar

[ref64] 64. Fehér B, Folyi T, Horváth J. Attention decreases auditory N1 latency. Front Neurosci Conference Abstract: 13th Conference of the Hungarian Neuroscience Society. 2011.
View Article
Google Scholar

[208] View Article

[209] Google Scholar

[ref65] 65. Schoonen R, Gelderen A Van, Glopper K de, Hulstijn J, Simis A, Snellings P, et al. First Language and Second Language Writing: The Role of Linguistic Knowledge, Speed of Processing, and Metacognitive Knowledge. Lang Learn. 2003;53: 165–202.
View Article
Google Scholar

[211] View Article

[212] Google Scholar

[ref66] 66. Gurin I. Priemy rechevoi kompressii pri sinkhronnom perevode c russikogo yazyka an angliiskii [Speech Compression Techniques For Simultaneous Interpetation From Russian Into English]. (PhD Thesis). Moscow State University; 2009.

[ref67] 67. Pellegrino F, Coupé C, Marsico E. Across-Language Perspective on Speech Information Rate. Language (Baltim). 2011;87: 539–558.
View Article
Google Scholar

[215] View Article

[216] Google Scholar

[ref68] 68. Simon SS, Tusch ES, Holcomb PJ, Daffner KR. Increasing Working Memory Load Reduces Processing of Cross-Modal Task-Irrelevant Stimuli Even after Controlling for Task Difficulty and Executive Capacity. Front Hum Neurosci. 2016;10: 1–13.
View Article
Google Scholar

[218] View Article

[219] Google Scholar

[ref69] 69. Sörqvist P, Rönnberg J. Individual differences in distractibility: An update and a model. PsyCh J. 2014;3: 42–57. pmid:25632345
View Article
PubMed/NCBI
Google Scholar

[221] View Article

[222] PubMed/NCBI

[223] Google Scholar

[ref70] 70. Bartłomiejczyk M. Strategies of simultaneous interpreting and directionality. Interpreting. 2006;8: 149–174.
View Article
Google Scholar

[225] View Article

[226] Google Scholar

[ref71] 71. Chernov G V. Message redundancy and message anticipation in simultaneous interpreting. In: Lambert S, Moser-Mercer B, editors. Bridging the Gap: Empirical research in simultaneous interpretation. Chapel Hill: John Benjamins Publishing Co.; 1994. pp. 139–153. https://doi.org/10.1075/btl.3.13che

Figures

Abstract

Introduction

Methods

Participants

Procedure and materials

Working memory load estimation

Transcript time coding and WM load estimation

EEG data acquisition and pre-processing

Data analysis

Results

WM load estimates from behavioral data

EEG results

P1 amplitude analyses.

N1 amplitude analyses.

P1 and N1 latency analyses.

Discussion

Limitations and future directions

Supporting information

S1 Appendix. Parameters of speeches used in the interpreting task.

S2 Appendix. The order in which the speeches were presented.

References