1.1 Aim

The aim of this study is to investigate the working memory capacity of dialogue interpreters. We start with the assumption that there may be a difference in certain central executive functions of the working memory, which can be traced to experience, as has been shown for simultaneous interpreting (see further below). We would also like to establish whether we can identify any differences in these central executive functions depending on the type of interpreter, dialogue interpreter vs. simultaneous conference interpreter. This leads to the following research questions:

1. Is there a difference in central executive functions (more precisely, attention control, updating, planning and resistance to interference) between experienced and inexperienced dialogue interpreters?

2. Is there a difference in these functions between dialogue interpreters and other types of interpreters (based on published results in other studies)?

2.1 Working memory in interpreting

Working memory as described by Wilhelm, Hildebrandt, and Oberauer (2013, p. 1) is a construct describing a system which gives access to information needed for ongoing cognitive processes (in our case, interpreting). Various models have been developed to describe this system, of which perhaps the Baddeley and Hitch's (1974) is the most well-known. Working memory capacity reflects the idea that there are limitations on individuals' working memory. In Interpreting Studies, the function or capacity of the interpreter's working memory (henceforth WM) is a well-researched topic. This research has mainly focused on simultaneous conference interpreters, and has sometimes involved comparing interpreters with bilinguals who are not interpreters. We list here several examples of important working memory studies related to interpreting. Liu (2001) investigated listening span, finding no difference between experienced interpreters and students, although there were differences in their interpreting performance. Timarová (2012) concluded in her PhD dissertation that WM is not a single entity in SI (simultaneous interpreting), that different WM executive functions have positive correlation to different simultaneous interpreting tasks, and that WM is a capability rather than capacity. According to her study, there are two relationships between WM and SI: one related to experience, the other to innate ability. The functions of dual tasking and resistance to interference are related to experience, while attention shifting, verbal memory and updating seem to be innate. Babcock (2015) concluded that SI is a skill that builds upon bilingualism. Verbal intelligence and verbal fluency are important positive predictive factors (which she does not label innate, but identifies as acquired before interpreting training). She found that interpreter training had an impact on several functions: verbal and spatial short-term memory, verbal working memory. Finally, she concluded that interpreters exhibit a smaller switching cost (the cost to switch between different tasks) compared to multilinguals without interpreting experience.

Lately, three meta-analyses on WM in SI have been published: Mellinger and Hanson (2019), Wen and Dong (2019) and Ghiselli (2022). Table 1 gives an overview of the studies included in the data-sets of the three studies.¹

The three meta-analyses are all done meticulously, overall following quite similar research procedures. The research questions posed differ to some extent, but are all concerned with interpreter advantage in WM, and the data analysed are from published empirical and/or experimental studies relevant to that question. It should be noted that, with the exception of Dong et al. (2018), all studies included in the data-sets of the three meta-analyses are on simultaneous interpreting.²³ The empirical data for the meta-analyses differs slightly in number: Mellinger and Hanson (2019) based their study on 16 publications, Wen and Dong (2019) on 10, and Ghiselli (2022) on 14 publications. Together, they cover 22 studies, published between 1995 and 2018. In view of the research questions posed, the criteria for inclusion in the meta-analyses differed slightly, which is one explanation for the different number of publications in the respective data-set; also, given the long publication times in many journals, it is natural that publications from 2018 could only be included in Ghiselli (2022). Only 3 studies (Köpke & Nespoulous, 2006; Signorelli et al.,2012;⁴ and Stavrakaki et al., 2012) were included in all three meta-analyses. Nine studies were included in two meta-analyses, and the rest, 10 studies, were included in only one of the meta-analytical studies (see Table 1).

The conclusions of the meta-analyses all point in the direction of an interpreter advantage in WM capacity, as shown by the following quotes: “…sufficient evidence seems to exist to declare that professional interpreters exhibit larger working memory capacities than non-interpreters” (Mellinger & Hanson, 2019, pp. 181–182); “…for the issue of an interpreter advantage (comparing interpreters with general bilinguals), an interpreter advantage was present in both WM and STM [short term memory] spans” (Wen & Dong, 2019, p. 778). Ghiselli (2022) did not specifically test the issue of the interpreter advantage as such, but focused on the types of tasks and what results they yield. Ghiselli (2022, pp. 71–72) could partially confirm greater differences in verbal tasks between study and control groups, but could not confirm that expertise has an effect on foreign language tasks, nor that tasks based on auditory stimuli (i.e., potentially more similar to interpreting) yield greater differences between study and control groups. She also mentions the fact that small sample sizes make reaching firm conclusions more difficult.

Furthermore, the meta-analyses conclude that the strength of the interpreter advantage is moderated by the type of tasks included in tests of WM, with verbal memory tests expressing the advantage more strongly: “This study found that interpreters and interpreting students have a significant WM advantage of medium size over non-interpreters in tasks based on verbal stimuli, but not in tasks based on non-verbal stimuli” (Ghiselli, 2022, p. 24). This is also noted by Wen and Dong (2019, p. 778): “…this advantage was more expressed in verbal memory tasks than in nonverbal ones”. Mellinger and Hanson (2019, p. 170) considered two types of tests/tasks from a different angle: “(1) memory tests involving storage only (short-term memory) versus those requiring cognitive processing (working memory); (2) the difference between visual and auditory stimuli”, and concluded regarding the second type that “modality of presentation [i.e., visual or auditory] does not have a substantial effect on the results of tests of working memory capacity. Professional interpreters outperform comparison groups, regardless of modality” (p. 182). Ghiselli (2022, p. 3) set out to test this conclusion, hypothesizing that tasks based on auditory stimuli would detect more differences between interpreters and control groups, but this hypothesis was not confirmed (p. 23).

The interpreter advantage is seen when comparing interpreters with control groups of non-interpreters. Furthermore, the meta-analyses analysed the data-sets for possible effect of length of interpreting experience on the results. Different approaches were used to operationalize the construct of interpreting expertise: Mellinger and Hanson (2019) used performance quality in simultaneous interpreting, Wen and Dong (2019) and Ghiselli (2022) used length of interpreting experience. Mellinger and Hanson (2019, p. 184) concluded that “…there appears to be a relationship between working memory capacity and the quality of simultaneous interpreting.” In the same vein, Wen and Dong (2019, p. 778) concluded: “our meta-analysis found that the level of interpreter expertise significantly moderated the presence or magnitude of such an advantage.” Ghiselli (2002, p. 23) reported similar findings.

Interestingly, Wen and Dong (2019, p. 779) report a lack of significant difference in WM and STM span between intermediate and expert interpreters, which tallies with the findings of Tiselius (2013) with regard to performance quality. Tiselius investigated the quality of three simultaneous interpreters' performance over time by assessing their interpreting of the same speech on two different occasions. Their performance was assessed by independent raters on two scales, intelligibility and informativeness. The study could not establish significant differences in the quality of their interpreting product recorded just before graduation and the one recorded after 15 years of experience. Although Tiselius' study was not directly related to WM, and of course needs to be replicated, it may indicate the presence of a threshold in development at the intermediate stage.

All three meta-analyses also consider the so-called publication bias, i.e., the fact that scientific journals prefer to publish studies that show statistically significant results, and that authors therefore may tend to refrain from submitting non-significant results for publication (see Rosenthal, 1979; Rothstein, Sutton, & Borenstein, 2005). This of course makes these results unavailable for meta-analysis, and may lead to a skewed analysis. The three studies took steps to analyse and report any publication bias; their findings indicate that publication bias is only a minor problem in studies on WM in interpreting.

This overview has shown that the WM capacity of simultaneous interpreters has been the subject of a number of experimental studies and three thorough meta-analytical publications with partly differing perspectives but very similar conclusions, regarding

1. the existence of evidence of an interpreter advantage in WM – at least in simultaneous interpreting;

2. verbal tasks demonstrating this advantage more expressly;

3. the claim that more experience in simultaneous interpreting gives a greater interpreter WM advantage, although some findings indicate that there is a threshold in the development of this advantage, which is reached at an intermediate stage of interpreting experience, and not passed with more experience.

In view of these results, we find it very relevant to test the WM capacity of spoken language interpreters who mainly work in the short consecutive mode. One such category is dialogue interpreters, working for instance in the public service sector. As far as we have been able to establish, there has not been any study on WM functions in dialogue interpreters.

2.2 Dialogue interpreting and cognition

We define dialogue interpreting as short consecutive interpreting (often utterance by utterance), with two or more participants in one conversation. In dialogue interpreting the interpreter works in both directions; this type of interpreting is common in the public sector (for instance medical encounters, job counseling). Simultaneous elements may also occur (cf. sign language interpreting). Dialogue interpreting has been and is often studied using conversation analysis (cf. Wadensjö, 1998).

In dialogue interpreting, the dialogue interpreter perceives an utterance in one language and renders it in its entirety in another language. The cognitive processes involved in this are related to perceiving the utterance, understanding it, storing it and recalling it for transfer into the other language, and producing it in the other language. These processes can take place both concurrently and/or consecutively. Just like in simultaneous conference interpreting, this involves simultaneous processing of utterances in two languages, and also constant switching between the two languages.

WM is involved in executing, monitoring and coordinating these processes, as well as problem solving. However, an important difference between the cognitive processing of dialogue and simultaneous interpreting is that simultaneous interpreters speak and listen simultaneously, which normally is not the case in dialogue interpreting. WM is also needed to monitor one's own interpreting process, taking into account cognitive constraints as well as the talk and interaction of other participants (Tiselius & Englund Dimitrova, 2023). The dialogue interpreter furthermore has limited opportunities to reduce the cognitive load through, for instance, taking notes or simultaneous production. In view of this, we find it reasonable to assume that dialogue interpreters will show an advantage in their WM capacity similar to that already evidenced for simultaneous interpreters. This assumption is strengthened by the results of Dong et al. (2018), who carried out a longitudinal study of WM capacity in students of consecutive interpreting, comparing them with a control group of L2 learners who had not been given interpreter training. Their study shows the importance of the updating function of WM, which is necessary to screen and code information and subsequently replace it with new information, connected to the function of storing the incoming utterance and recalling it for transfer and production. They conclude that “…updating efficiency (even when measured by a non-verbal version of the n-back task as in the present study) is more central to the CI [consecutive interpretation] task than WM size, at least for beginning student interpreters like ours, and is therefore more exercised in CI training, leading to an interpreter advantage in updating efficiency” (Dong et al., 2018, p. 10).

2.3 Hypotheses of the study

In order to investigate our research questions, we formulated the following two hypotheses:

1. There will be measurable significant differences in WM tests between experienced and inexperienced dialogue interpreters, and

2. There will be no significant differences in WM tests between experienced dialogue interpreters and experienced simultaneous conference interpreters.

3.1 Method

Participants were on site at university premises, and completed the whole data collection process in one session. The psychometric test battery was the initial part of the data collection for each participant. First, a general intelligence test, a so-called Raven test, was administered by pen and paper to all participants. After the Raven test, the participants completed computer-based WM tests with the aim of measuring executive functions. Since the data set of this study is drawn from a larger project with a mixed-methods design, the participants also took the Dialang test (Huhta, Luoma, Oscarson, Sajavaara, & Teasdale, 2002) and interpreted a simulated job counseling encounter after the working memory tests.

As WM is not directly observable, the tests were chosen with the aim of reflecting the components of WM, in which interpreters have previously shown an advantage as compared to bilinguals (see above), and in which we could possibly trace an effect of experience for dialogue interpreting. Given the desire to compare experienced dialogue interpreters with experienced simultaneous interpreters, some of the tests chosen were also the same as those used by Timarová (2012). The tests chosen covered the following areas:

1. verbal memory, which is memory involved in identifying and remembering words and other abstract entities related to language. The tests chosen were:

   1. a letter span task. The test consists of strings of individual letters that the participants see and are then asked to recall in the order the letters were presented.

   2. an operation span task. In this test, the participants were first shown three to seven consonants (randomly generated by the software out of 12 possible), then told to perform an intervening arithmetic operation (e.g., (2 × 6) - 4 = ?), and then asked to recall the consonants shown before the arithmetic operation.

2. visuospatial memory, which is needed for keeping track of the location of moving objects in order to remain oriented in space. The test chosen here was:

   1. the matrix span, where the task is to recall the position of two to seven red squares in a 4 × 4 grid.

3. updating, which is the WM function allowing us to screen and code information, and then replace it with new information. It is relevant for instance for decoding and remembering verbal information or anticipating information. The test chosen was

   1. the 2-back test, which consisted in showing the participants a series of letters, one by one, and then asking them to judge whether the currently shown letter is identical to the one shown two steps back. The participants therefore have to constantly hold in memory the two previous letters in order to fulfill the task.

4. dual tasking and attention switching, which require a lot of general WM resources and also reflect processing capacity limits. These are skills that interpreters presumably possess or develop. The test chosen here was

   1. the Barrouillet test (Lépine, Bernardin & Barrouillet, 2005). The Barrouillet test is similar to the operation span task (string of letters + arithmetic operation or parity judgment), but the Barrouillet test also takes increased attention load and its effect on memory task performance into account. This is created by increasing the speed at which the tasks are presented.

Participants also took the arrow flanker test, which is a test of resistance to interference. This is a test where the participants' ability to ignore irrelevant stimuli is tested, a skill probably relevant for interpreting. However, due to a malfunction the data for this test was lost.

As shown above, these tests cover several different aspects of WM. We have in particular been interested in areas that would be more relevant to show the effect of interpreting expertise. As already mentioned, some of the tests (namely the letter span, the 2-back and the Barrouillet) were chosen in particular to be able to compare the results of our experienced interpreters to those of Timarová’s (2012) study on WM. Timarová studied 28 conference interpreters with Dutch (n = 8) or Czech (n = 20) L1 and with English as one of their working languages.

The test was administered and answered via computer, but for the letter span test and the Barrouillet test the participants responded (i.e., noted down the letters they were asked to recall) with pen and paper. In order to analyse the data from these tests, the researchers first had to feed the answers from the response sheets into a raw data excel file. After this primary preparation of the data, the results from the tests were analyzed statistically, and the difference between the different groups of participants were checked using a t-test.

3.2 Participants

Fifteen interpreters with different levels of experience and different language backgrounds participated in the study. All participants had Swedish as one of their working languages.

As can be seen from Table 2, there were eight experienced participants, all of whom had Swedish in combination with either French, Spanish or Polish. All eight were state authorized with at least five years of working experience. The remaining seven participants were inexperienced interpreters who had Swedish in combination with one of the three just-mentioned languages. The inexperienced interpreters were either students in their last term (of three) at Stockholm University's full time bachelor course in public service interpreting (n = 5) or had completed a folk high school programme⁶ and had one year of interpreting experience (n = 2).

Participants were found via recruitment in the interpreting programmes at Stockholm University (inexperienced interpreters), and via e-mail sent to Swedish state authorized interpreters with the desired language combinations (experienced interpreters). The recent graduates from the folk high school programme were recruited through personal contacts. Participants were paid for their time. Students were paid the standard amount used at university for student jobs, and professionals were paid the same hourly rate as for interpreting an hour in court. The total remunerated number of hours per person was six. All participants were informed about the project and signed an informed consent form before starting the data collection. The project was given ethical approval by the ethical review board of the Stockholm region (2017/2208-31).

The participants used as a comparison group in order to test hypothesis 2 were those recruited for Timarová's (2012, p. 92) study. She recruited 28 participants (18 female and 10 male) for her study, all of whom were conference interpreters. The inclusion criteria were being an active professional interpreter accredited to work at EU institutions with L1 either Czech (n = 20, 5 male) or Dutch (n = 8, 3 male) and with English in the language combination. Participants had a university diploma and were between 25 and 55 years of age. They had between 1 and 25 years of professional practice, the average being 11.9 years (SD = 6.9 years).

4.1 Experienced and inexperienced dialogue interpreters

The results from the Raven test can be found in Table 3. The p-value for the difference between the two groups was 0.46 (t = 0.765, df = 13, std err = 0.732) and thus not significant.

Tables 4 and 5 give the results from the verbal memory tests. Table 4 shows the result from the letter span test. The p-value for the difference between the two groups was 0.82 (t = −0.231, df = 13, std err = 0.173) and thus not significant.

The results from the operation span tests are shown in Table 5. Unfortunately, for this test, data from two participants was lost. The p-value for the difference between the two groups was 0.27 (t = −1.155, df = 11, std err = 0.13) for the operation score and 0.74 for the (t = −0.339, df = 11, std err = 0.089) operation total. No significant difference could thus be established.

For visuospatial memory, results from the matrix span test are shown in Table 6. The p-value for the difference between the two groups was 0.51 (t = −0.682, df = 13, std err = 0.088) for the matrix score and 0.88 (t = −0.147, df = 13, std err = 0.068) for the matrix total. Hence no significant difference could be established.

In Table 7, results from the 2-back test, measuring updating capacity, are shown. The p-value for the difference between the two groups was 0.83 (t = 0.215, df = 13, std err = 0.093) and thus no significant difference was established.

Table 8 shows the results of the Barrouillet test, on dual tasking and switching. Barrouillet has a fast recall task and a slow recall task. In Table 8, the average of both tests is given. For the Barrouillet test there was also a data loss from two participants. However, once again the p-value for the difference between experienced and inexperienced participants was not significant (p = 0.59, t = 0.561, df = 11, std err = 0.160).

The results above are all related to our first hypothesis, which stated that we would find measurable significant differences in WM tests between experienced and inexperienced dialogue interpreters. This hypothesis was not confirmed, we found no significant differences between experienced and inexperienced dialogue interpreters in any of the collected test data.

4.2 Experienced dialogue interpreters and experienced simultaneous interpreters

We next present a comparison between Timarová's experienced simultaneous conference interpreters and the experienced dialogue public service interpreters who participated in this study. The comparison with Timarová is done with the WM tests for letter span (verbal memory), 2-back (updating) and Barrouillet (general WM) (Timarová, 2012, p. 94, Table 5.2).

Table 9 shows the results for the verbal memory test. The average portion of correct answers for the conference interpreters was 0.63 and for dialogue interpreters it was 0.46. This difference was significant with a p-value of 0.0317 (t = 2,240, df = 34, std err = 0.076).

In Table 10, we give the results for the portion of correct answers in the 2-back test measuring updating capacity. Timarová’s conference interpreting participants had a proportion of correct answers of 0.89 and our dialogue interpreting participants' proportion of correct answers was 0.74. This difference is significant with a p-value of 0.0006 (t = 3.782, df = 33, std err = 0.040).

Finally, Table 11 shows the result of a general WM test, i.e., the Barrouillet test. The figures show the average of the two conditions in the test, viz. fast recall and slow recall. In the Barrouillet test the simultaneous interpreters' mean is 0.98 and the dialogue interpreters' mean is 0.94. This difference is not significant, with a p-value of 0.61 (t = 0.522, df = 32, std err = 0.077).

Our hypothesis 2 stated that we would not find significant differences in WM tests between experienced dialogue interpreters and experienced simultaneous interpreters. To test it, we compared our experienced participants with simultaneous conference interpreters in Timarová’s study. As can be seen in Tables 9 and 10, the participants in Timarová’s study outperformed the experienced dialogue interpreters in our study. There were significant differences in the test results for verbal memory, updating, visuospatial memory between the dialogue interpreters in our study and the conference simultaneous interpreters in Timarová’s study. However, for the general working memory test (Barrouillet), Table 11, there were NO significant differences in the test results between the two groups.