On the encapsulation of bilingual language control

https://doi.org/10.1016/j.jml.2018.12.001Get rights and content

Highlights

  • Bilingual advantages in inhibitory control did not occur in four tasks.

  • Inhibitory control was not affected by specific aspects of bilingual experience.

  • Interference tasks did not cluster based on stimulus–response compatibility.

  • Spatial stroop and Simon tasks showed significant inter-task correlations.

Abstract

One purpose of this study was to test the hypothesis that prevalent patterns of bilingual language control lead to greater enhancement of the ability to resolve Stimulus-Stimulus conflict compared to Stimulus-Response conflict. To that end 104 bilinguals and 62 monolinguals completed four commonly used nonverbal interference tasks with varied S-S and S-R incompatibilities. No bilingual advantages were observed in any of the tasks. A second purpose was to further investigate whether a general inhibitory-control ability exists by examining inter-task correlations in the current and previous studies. We conclude that there may be a shared mechanism for interference control across spatial Stroop tasks and Simon tasks, but that this mechanism is apparently not recruited during bilingual language control to the extent that such practice would enhance a general ability. Rather, inhibition in language processing may be encapsulated within a lexical-identification system as assumed by Dijkstra and van Heuven’s bilingual-interactive-activation plus model and its update Multilink.

Introduction

Bilinguals have been claimed to perform better than monolinguals in nonverbal interference tasks because they constantly practice inhibiting the language currently not in use. There is no doubt that when bilinguals intend to produce a sentence in a target language, the translation equivalents in the other language become coactivated and create competition (see Paap, 2018 for a review). For example, the intention to say “bull” may coactivate “toro” in a Spanish-English bilingual. Likewise, during bilingual language comprehension similar or identical word forms are coactivated in both mental lexicons. For example, even in an English context /lief/ activates the word for “sweet” (a false cognate) in Dutch-English bilinguals. There are several possible adaptations to this competition: (1) the non-target lexical competitors are inhibited via a general inhibitory control mechanism, (2) the non-target lexical competitors are inhibited via a specialized mechanism within a word identification system, (3) the target candidates are up regulated, or (4) no conflict resolution mechanism is employed at the level of lexical representations. If the last is true, then bilinguals (5) may live with the competition and the occasional unintended intrusion from the other language, (6) employ domain-general response suppression, or (7) rely on a specialized mechanism for articulatory suppression. Some of these options could occur in combination, but logically only when (1) or (6) are involved could bilingual language-control transfer to and lead to bilingual advantages in nonverbal interference tasks requiring manual responses.

There are influential models of bilingual language-control that either assume or eschew the recruitment of general inhibitory control. Green’s (1998) inhibitory control model (ICM) presumes the need for higher-level task schemas controlled by an even higher level supervisory attentional system (SAS). For example, when presented with a printed word a bilingual can be given the task of reading it silently, reading it aloud, translating it, generating an associate, classifying it as animate or inanimate, and so on. Each requires a different task schema. The ICM emphasizes that task schemas compete with each other for controlling action. Thus, depending on changes in topic or conversational partners, bilinguals may decide to switch from the “speak-English” schema to a “speak-Spanish” schema that reactively inhibits the English lexical representations via their language tags. The crucial point is that bilingual language control is governed by a SAS that selects and schedules specific task schemas (e.g., “speak English”) in the same way as it does in novel situations and tasks (e.g., “name the color” in the standard Stroop task). If the ICM assumptions are correct, then bilingual advantages should frequently occur in nonverbal tasks where conflict between stimulus representations needs to be resolved.

A contrasting view is presented by Dijkstra and van Heuven (2002) bilingual interactive-activation-plus model (BIA+). The BIA+ makes a clear distinction between the encapsulated word identification system (WIS) and domain-general cognitive control (the task schema/decision system). Because the WIS is encapsulated, the activation of the word-form nodes cannot be affected by top-down control. The WIS is assumed to be part of a larger language processing system. The BIA+ further assumes that the syntactic and semantic constraints of this language processor can affect the activation of the lexical nodes. In the BIA+ the activation of word-form units is affected only by processing within the language processing system. Linguistic information from the visual input (c-a-t) or from the preceding sentence context (e.g., The mouse was chased by the…) can and will affect the activation of word form units, but nonlinguistic context such as the participant’s expectations and strategies will not. Rather, these nonlinguistic factors influence performance via adjustments to parameter settings in the task schema/decision system. The decision mechanism is part of a task schema and reads out the activation of nodes within the WIS.

For example, in general lexical decision tasks a word response can be made when a word unit from either language crosses an activation threshold. As the task unfolds, the activation thresholds and temporal deadlines may be adapted based on the frequency of different types of items. These adaptations involve dynamic adjustments of response criteria, but do not affect the activation level of individual word form units or the language nodes. For present purposes, the crucial aspect of BIA+ is that inhibitory control is domain specific and encapsulated within the word-identification and language processing system. It is a specific instantiation of Adaptation 2 from above that the non-target lexical competitors are inhibited by a specialized mechanism. If the BIA+ correctly characterizes bilingual language control, then there should be no differences between bilinguals and monolinguals in nonverbal tasks requiring conflict resolution between stimulus representations because task-specific mechanisms, by definition, do not generalize.

The computational version of the BIA+ was limited to visual word recognition, but as a verbal model included both phonological and semantic representations. Dijkstra et al. (2018) extended the computational version of BIA+ to production. This new model, dubbed Multilink, successfully simulated many classic results obtained with picture naming and translation tasks. Critical to the present discussion, bilingual language control remains encapsulated within the WIS and thus the model predicts no transfer to tasks using general executive functioning (EF). Critical to an issue raised later, the WIS eschews inhibitory connections both within and between representational levels. This is an instantiation of Adaptation 4 that no conflict resolution mechanism is employed at the level of lexical representations. To state the obvious, if there are no inhibitory connections, then there is no inhibitory control ability that can be enhanced by practice.

Although the ICM and BIA+/Multilink make clear and opposing assumptions regarding the direct application of general inhibitory control on lexical units, the models do not explicitly consider whether or how conflict might be resolved at the response level. To return to the earlier example, if the orthographic word forms for both BULL and TORO are coactivated when the intention is to read aloud in English, but the lexical competitor TORO is not suppressed, then the articulatory plans associated with the corresponding phonological word forms will compete. Inhibition may be applied in order to induce fluent speech and avoid intrusions from the unintended language. If this recruits a general inhibitory control mechanism, then that mechanism may be strengthened through ubiquitous practice.

The possibility of inhibitory control at either the stimulus or response levels has been extensively studied in cognitive psychology. Consider the most influential taxonomy for analyzing task differences, Kornblum (1994) Dimensional Overlap Model. The model distinguishes between tasks with stimulus-response (S-R) or stimulus-stimulus (S-S) incompatibility. The incompatibilities of the four tasks used in the present study are illustrated in Fig. 1. For each panel the S-R rule is at the top, a display representing a correct response on an incongruent trial is in the middle, and the Venn diagrams at the bottom represent, by their intersections, where conflict can be generated and resolved. The first (leftmost) panel is a pure S-R task that is often referred to as a Simon task. A single arrow (pointing either up or down) is presented either to the left or right of fixation and the rule is to press the left key if the arrow points up and the right key if it points down. Given the natural tendency to react toward the source of stimulation (Simon & Small, 1969), competition can occur when the physical location and the rule are incongruent as illustrated by overlap between SI (the irrelevant stimulus) and R (the correct response). Note that there is no overlap between the task relevant and task irrelevant stimulus representations because the arrow’s form varies on an up-down dimension whereas its location varies on a left-right axis. More generally, on the incongruent trials of an S-R task the response specified by the task rule is incompatible with the response of a prepotent but task irrelevant stimulus.

If the vertical arrows are displaced either above or below fixation as in third panel, the task transforms into a pure S-S task. Because the upward pointing arrow appears below the fixation, the task-relevant dimension (up arrow) is opposite its location (below) and causes S-S incompatibility. There is no S-R incompatibility because the layout of the response keys (horizontal) is orthogonal to the up-down direction of the arrow. This pure S-S task will be referred to as the vertical Stroop task.

As illustrated in the second panel tasks have both S-S and S-R incompatibilities when arrows pointing left or right are displaced either to the left or right of fixation. On an incongruent trial the task-relevant direction (e.g., left) is incompatible with task-irrelevant location (e.g., right) causing S-S incompatibility. Furthermore, the task-relevant direction (e.g., left) is also incompatible with the predisposition to react toward the source of stimulation (e.g., right) causing S-R incompatibility. This task will be referred to as the spatial Stroop task and in the literature is sometimes referred to as the Simon arrows task.

The rightmost panel illustrates a variant of the flanker task. If the central arrow points left press the left key, if it points right then press the right key. The flanking arrows are irrelevant but when they point in the opposite direction there is both S-S and S-R incompatibility because the relevant and irrelevant stimuli share the same dimension (viz., left-right) as do the relevant stimulus and the response.

The Kornblum taxonomy and this set of tasks provides a testbed for how S-S or S-R inhibitory control is moderated by any individual difference or group variable such as aging, video gaming, or bilingualism. Given bilingualism as our focus, here are some possibilities. If S-S conflict is resolved by language-specific mechanisms (as assumed in the BIA+/Multilink), then no bilingual advantage will accrue due to S-S incompatibilities and no advantage will occur in the vertical Stroop task as its only source of conflict is S-S. In contrast, if domain-general inhibitory control is routinely recruited (as assumed in the ICM), then bilingual advantages would be expected in the three tasks that have S-S incompatibilities. To take a third example, if a general response suppression mechanism is routinely recruited, during bilingual language control, then a bilingual advantage should appear in the three S-R tasks. Finally, if bilingual language control does not involve general inhibitory control at either the lexical or response levels, then there should be no performance differences between bilinguals and monolinguals on any of the tasks. The consequences of other permutations can be deduced from inspection of Fig. 1.

Blumenfeld and Marian (2014) developed a specific hypothesis based on the Kornblum taxonomy and the following analysis of bilingual language control. They reasoned that S-S competition may be the most common type of bilingual competition because competition between the lexicons occurs during both comprehension and production, whereas S-R inhibition may be limited to production contexts where both languages remain active until the response stage. Because cross-linguistic co-activation results in S-S competition more frequently than S-R competition, they predicted that bilingual advantages would be larger in an S-S task compared to a pure S-R task. The S-S task tested by Blumenfeld and Marian was like the spatial Stroop task illustrated in the third panel of Fig. 1 and their S-R task was like the Simon task shown in the first panel.

Across Blumenfeld and Marian’s two experiments there was some evidence for bilingual advantages in the spatial Stroop task. Experiment 1 recruited young adults from the Chicago area and included 38 English-Spanish bilinguals and 30 English monolinguals. Experiment 2 recruited young adults from San Diego with 60 participants in each group. Overall accuracy (about 90%) across experiments and groups was somewhat lower than typically observed for young adults in nonverbal interference tasks (see Table 1). To protect against a speed-accuracy trade-off Blumenfeld and Marian used efficiency scores (ES) as a composite measure of speed and accuracy. An ES is calculated as the mean correct RT divided by the proportion correct (PC). In Experiment 1, the Task × Group × Congruency interaction was significant for the ES and PC measures but not for RT. For the ES and PC measures, the pattern of interaction was as predicted: larger bilingual advantages in the S-S task and larger task differences for bilinguals. Despite the larger sample size and greater power, neither the Group × Congruency nor the three-way interaction with Task was significant for any of the three measures in Experiment 2 and consequently there was no statistical support in the second experiment for the hypothesis that the interference effects would be smaller in bilinguals compared to monolinguals in the S-S (spatial Stroop) task. Blumenfeld and Marian’s reconciliation of the inconsistencies between their two experiments is considered in our general discussion.

Blumenfeld and Marian reviewed 21 prior tests and showed that bilingual advantages tend to occur more often in S-S tasks compared to S-R tasks. Only one pair of tests in their review involved the same sample of participants. Although their review supported the hypothesis that bilingual advantages occur more often in S-S tasks it is risky to compare the results of studies using an S-S task to different studies using an S-R task. A chronic problem in testing for bilingual advantages is that bilinguals and monolinguals often differ in terms of ethnicity, culture, education, immigrant status, SES and other factors that may influence measures of inhibitory control (see Paap, Johnson, & Sawi, 2015, for a review). For that reason, it is better to focus on studies that test the same language groups with both an S-S and an S-R task. In these studies, any confound that favors one language group over the other should apply a comparable bias to each task.

Do bilingual advantages occur more often in S-S tasks when the same participants perform both types of tasks? Table 1 shows the results of each such study and a summary on the bottom row. The first important observation is that statistically significant advantages in inhibitory control are infrequent. There was only 1 bilingual advantage and that occurred in a pure S-R Simon task. There were also 4 monolingual advantages, and 19 non-significant differences. When the magnitude of the bilingual advantage is averaged across studies, neither task type shows a bilingual advantage, and the magnitude of the monolingual advantage (−6.5 ms) is actually greater for the S-S tasks than for the S-R tasks (−3.9 ms). In summary, when examining only those studies in which the same participants completed both an S-S and S-R task the evidence supporting the hypothesis that bilingual-language control employs a general S-S inhibitory control mechanism is absent.

To this point the emphasis has been on whether S-S tasks may yield more consistent bilingual advantages than S-R tasks. Blumenfeld and Marian’s review of mostly between-subject comparisons appeared more promising in that regard than our review of the 12 studies that used within-subject designs. Given that the S-S versus S-R distinction may not be a good predictor of bilingual advantages, it is also instructive to look at all of the data derived from nonverbal interference tasks that include both congruent and incongruent trials. Three recent meta-analyses converge on the conclusion that significant bilingual advantages in inhibitory control are relatively rare (15% of all comparisons), that the average effect sizes are very small, and that there is evidence for publication bias, which when taken into account, appears to completely eliminate the effect. In the meta-analysis by Paap (2019) the mean bilingual advantage across all 146 comparisons was +4.4 ms. If the 146 effect sizes are treated as a single sample the Bayes Factor (using the JZS prior and Rouder’s calculator) favoring the alternative is 2.9, an odds ratio that according to Jeffreys (1961) guidelines is barely worth mentioning. The meta-analyses by Lehtonen et al. (2018) examined bilingual advantages across six domains of EF, but their analysis of inhibitory control is most central to our focus. Their meta-analysis used a wider definition of inhibitory control tasks and identified a more heterogeneous set of 212 effect sizes compared to Paap (2019). The mean effect size for inhibitory control in Lehtonen et al. was Hedge’s g = +0.11 [+0.05, +0.18], but when corrected for bias the mean was no longer significant, g = −0.02 [−0.12, +0.08]. Donnelly, Brooks, and Homer (in press) reported a meta-analysis of 80 studies using a multiverse analysis approach where each research question was tested many times while making different decisions about the inclusion criteria. The bilingual-advantage effect size, corrected for publication bias, was negative, g = −22 [−0.35, −0.09]. The Lehtonen et al. meta-analysis was restricted to studies using participants 18 years and older, Paap to 6 years and older; and Donnelly et al. to 4.5 years and older.

If bilingual advantages in inhibitory control are indistinguishable from zero, then neutral observers, and certainly skeptics, may well question the value of another study comparing bilinguals to monolinguals. However, proponents of the bilingual advantage hypothesis continue to argue that bilingualism enhances domain-general EF, but only when sufficient dimensions of bilingual experience are combined with sensitive tests of the affected component(s) of cognitive control (e.g., Bialystok, 2017, Tabori et al., 2018, Struys et al., 2018). For that reason it is worthwhile to conduct a study that compares four closely-matched tasks that differ with respect to the need for spatial attention (only the flanker does) and includes pure S-S (vertical Stroop) and pure S-R (Simon) tasks.

On the other hand the possibility that bilingual advantages in inhibitory control do not exist raises the question why? One explanation already introduced is that bilingual language control relies on language-specific, not domain general, control mechanisms. That is, bilingual language control may be encapsulated within the language-processing system as implemented in the architecture of Multilink. A related, but more expansive explanation is that domain-general and top-down inhibitory control may not exist, that is, inhibitory control may be task-specific. Thus, an additional important purpose of the present study is to determine the psychometric structure among a set of four tasks that are frequently used to measure “inhibition”.

If the interference scores derived from any two nonverbal interference scores correlate, this can be taken as evidence that they share a conflict-resolution mechanism. However, Kornblum’s taxonomy implies that different mechanisms are employed to resolve S-S and S-R conflict. Thus, the intertask correlations between the interference scores for the four tasks shown in Fig. 1 should increase as pairs of tasks share neither type of conflict, one type, or both.

Only three of the six possible pairs from this set of four tasks have been previously tested using within-subjects designs. Each of the three pairings are discussed next. The interference scores for the pure S-R Simon task and the pure S-S vertical Stroop should not correlate if they rely on different inhibitory abilities. Although nine published studies had the same participants do both tasks (either as separate tasks or as an integrated task with trial types mixed), eight did not report intertask correlations. However, one study did (Li, Nan, Wang, & Liu, 2014) and the outcome was consistent with the prediction that S-S and S-R tasks employ separate mechanisms. They reported a small, negative, and non-significant correlation, r(28) = −0.05 between a pure S-R Simon task and a pure S-S vertical Stroop task. This finding is shown in the top block of Table 2.

As shown in the second block of Table 2, 12 experiments have correlated a pure S-R Simon task with a flanker task that includes both S-S and S-R incompatibilities. The intertask correlations between the interferences scores for all 12 were nonsignificant with Pearson r’s ranging −0.02 to +0.23. A parsimonious interpretation is that the pure S-R Simon task and the flanker task do not share a conflict resolution mechanism. Thus, the potential S-R conflict in a flanker task does not appear to be sufficient to generate a correlation with the interference scores in a pure S-R Simon task.

Based on the Kornblum taxonomy, the cases most likely to correlate have paired the spatial Stroop task with a flanker task as both tasks include both S-S and S-R incompatibilities. However as shown in the third block of Table 2, neither de Bruin and Della Sala (2017) nor Pettigrew and Martin (2014) found a significant correlation. In contrast, for the experiment described in Wöstmann et al. (2013) there was a significant correlation (U. Ettinger, personal communication, r(534) = +0.24, p < .001). The effect size is in the small to moderate range and is highly significant, in part, because of the large degrees of freedom.

The overall pattern of intertask correlations is suggestive, at best, that nonverbal interference tasks involving S-S incompatibilities recruit a common inhibitory control mechanism. If they did (and if bilingual language control also involves the same ability to resolve S-S conflict), then the three S-S tasks should produce bilingual advantages. However, given that the intertask correlations are mostly small and nonsignificant, the outcomes reviewed in Table 1 are also consistent with the possibility that inhibitory control in S-S tasks is task specific and that no bilingual advantages should be observed.

Exploring the construct of inhibition through individual differences often goes beyond the zero-order intertask correlations and uses latent-variable analyses such as confirmatory factor analysis (CFA) and/or structural equation modeling (SEM). This approach has not been used to examine S-S versus S-R conflict resolution as such, but in their seminal work Friedman and Miyake (2004) considered very similar constructs dubbed inhibition of prepotent responses1 (i.e., the ability to suppress dominant responses) and resistance to distractor interference (i.e., the ability to ignore distracting information). Inhibition of prepotent responses is more closely aligned to S-R incompatibility and resistance to distractor interference to S-S incompatibility. The CFA indicated that the two types of inhibition could be considered separate but correlated factors. However, additional modeling using SEM led Friedman and Miyake to prefer a more parsimonious model that collapsed the initial two factors into a single latent variable.

The question of whether inhibition of prepotent responses (akin to S-R conflict) and resistance to distractor interference (akin to S-S conflict) should be considered separate factors was further and more extensively examined by Rey-Mermet, Gade, and Oberauer (2018). Measures from six tasks were assumed to load on inhibition of prepotent responses and four others on resistance to distraction. The results were analyzed and interpreted in two distinctively different ways. The competitive fitting of SEM models (the standard approach) yielded a best-fitting model with two positively correlated inhibition factors assumed to reflect inhibition of prepotent responses and resistance to distractor interference. This analysis reinforces a common belief that inhibition is a useful psychometric construct and that it can be differentiated into correlated, but separable latent variables.

The alternative interpretation, and the one that Rey-Mermet et al. urge us to seriously consider leads to the conclusion that there is no general ability for inhibitory control. Here are the building blocks of their argument. First, the intertask correlations for measures in their study that were intended to load on the same factor were typically low. These low intertask correlations imply that most of the variance in the inhibition measures is not accounted for by the latent factors even when the overall fit of the models by conventional standards is good. Second, each factor was dominated by a single measure. They show this trend not only in their own data, but also in a fair amount of previous work (see Rey-Mermet et al.’s Table 8). Rey-Mermet et al.’s third argument is based on Bayesian hypothesis testing. These tests used a measure of model fit called the Bayesian information criterion (BIC) approximation (Wagenmakers, 2007). For example, the BIC for a model with only a single inhibition factor was cast in the role of the null hypothesis while the BIC for a two-factor model was cast in the role of the alternative hypothesis in order to compute a Bayes factor in favor of the null hypothesis (BF01) and in favor of the alternative (BF10). The advantage of using Bayesian hypothesis testing is that, in theory, one can not only reject the null hypothesis, but also accept it given that the BF is strong. However, the results were ambiguous. The BF did not distinguish between the single-factor and two-factor models. Neither did it distinguish between the two-factor model with correlated factors and a two-factor model with completely unrelated factors. Putting this together the BFs do not permit us to conclude whether there is one inhibition factor or two; or, if two, whether they are correlated or orthogonal. Furthermore, given that the factors tend to be dominated by a single measure and that the simple intertask correlations are very small, it is plausible that inhibition is task specific and not a general ability. Rey-Mermet et al. conclude their paper by suggesting that “we should perhaps stop thinking about inhibition as a general cognitive construct” p. 516.

If inhibition is task specific this would explain why inhibition sometimes fails to surface as a coherent latent variable that is separable from other factors. For example, in a CFA intended to confirm updating, shifting, and inhibition as factors, Krumm et al. (2009) found that the three presumed measures of inhibition (antisaccade, stop-signal, and a Stroop task) did not form a coherent latent variable. Similarly, van der Sluis, de Jong, and van der Leij (2007) explored the same three-factor model of EF by testing 172 children on 11 tasks that included the Stroop color-word naming task and three other Stroop variants. A common factor for inhibition could not be distinguished from the naming control tasks. In yet another attempt to confirm a three-factor solution of EF Hull, Martin, Beier, Lane, and Hamilton (2008) concluded that the inhibition factor was not adequately determined. Less stark results, but still amenable to there being no separable inhibition factors, comes from Klauer, Schmitz, Tiege-Mocigemba, and Voss (2010) who reported that the two measures (antisaccade and stop-signal) intended to load on an inhibition factor were better incorporated into the working memory factor. The absence of a separable inhibition factor is also consistent with Miyake and Friedman’s (2012) more recent unity and diversity model of EF. In multiple data sets the three measures selected to load on an inhibition factor were not separable from a basic EF ability. They explicitly reject the interpretation that the common EF factor is inhibition or that inhibition is the most central of all EFs. Rather, they suggest that the common EF factor reflects individual differences in the ability to maintain and manage goals and to use those goal to bias ongoing processing. In summary, the results of most latent-variable investigations could either neatly fit or stretch to fit a loss of faith in inhibition as a general construct.

For some readers abandoning inhibition as a general construct may seem a bridge too far, but Rey-Mermet et al. were not the first to cross it. In a highly cited essay titled “In Opposition to Inhibition” MacLeod, Dodd, Sheard, Wilson, and Bibi (2003) provide alternatives to inhibition for several classic demonstrations of “inhibition” (e.g., Stroop color-word interference, inhibition of return, negative priming, etc.). These alternative explanations describe how standard functional or neural net models might predict interference effects without having to evoke inhibitory control.

Hampshire and Sharp (2015) also concluded that top-down inhibitory control is not necessary. Using the stop-signal task as their primary example, they argued that top-down inhibition from frontal cortex to other areas (or even to specific representations) is simply not necessary because phenomena assumed to reflect top-down inhibition can be explained by the upregulation of correct responses that compete with incorrect responses through local lateral inhibition. They suggest that an illusion of top-down inhibition may have been created by overgeneralizing a real neural mechanism (local lateral inhibition) to one that has not been established (a global inhibitory mechanism acting between cortical regions). Similarly, and based primarily on cognitive neuroscience data derived during the antisaccade task, Curtis and D’Esposito (2013) proposed that cognitive control should be modeled as a process by which the best response (including not responding at all) among competing responses is selected. This led them to conclude that a mechanism specialized for inhibiting actions, per se, does not seem necessary for the behavioral expression of inhibiting an unwanted action.

The present study has two purposes. One is to determine the psychometric structure among a set of four tasks that are frequently used in cognitive psychology to measure “inhibition.” All four allow the computation of interferences scores (the difference between congruent and incongruent trials) and based on Kornblum’s taxonomy they include a pure S-S task (vertical Stroop), a pure S-R task (Simon), and two that involve both types of conflict (spatial Stroop and flanker). Predictions based on S-S and S-R overlap between tasks will be evaluated with intertask correlations and an exploratory factor analysis. Another important purpose is to test the hypothesis that bilingual advantages are most likely to occur on tasks that involve S-S conflict. This hypothesis is based on the assumption (e.g., Green’s ICM) that bilingual language control and the control recruited on nonverbal interference tasks with S-S conflict are one and the same. In contrast, the absence of bilingual advantages across all four tasks would, of course, be consistent with the encapsulation assumption of the BIA+/Multilink models.

Section snippets

Sequence of events

All parts of the study were conducted in a single session of at least 60 min.2 The first activity was obtaining written consent to participate using a form approved by the SFSU IRB. This was followed by: (1) the four nonverbal interference tasks, (2) the Raven’s test, (3) the MINT task of productive vocabulary in English, and (4) the background

Trimming of response latencies

Consistent with Blumenfeld and Marian’s statistical analysis, RTs less than 200 ms or more than 2.5 standard deviations above the participant’s mean were removed. However, given Zhou and Krott (2015) hypothesis that bilingual advantages are primarily driven by the right tail of the RT distribution all analyses were repeated trimming only RTs that were greater than three seconds. In our tasks, a trial did not terminate until there was a valid response and, consequently, a distracting event might

General discussion

If all four tasks had produced robust bilingual advantages this would imply that there is a general inhibitory control ability that also plays a prominent role in bilingual language control. Alternatively, if one of the tasks or a subset of the tasks had produced robust bilingual advantages this would imply that there are at least two conflict resolution mechanisms and that the tasks producing advantages employ the same mechanism that is prevalent during bilingual language control.

Conclusion

The first major purpose was to determine the psychometric structure among a set of four tasks that are frequently used to measure inhibition. The three tasks where conflict is between two dimensions of the same stimulus formed a coherent latent variable that excluded the flanker effect where the conflict is between different stimuli. This pattern was counter to predictions based on Kornblum’s taxonomy, but consistent with the assumption that selective attention to objects or spatial locations

References (66)

  • H. Magen et al.

    Modularity beyond perception: Evidence from single task interference paradigms

    Cognitive Psychology

    (2007)
  • K.R. Paap et al.

    There is no coherent evidence for a bilingual advantage in executive processing

    Cognitive Psychology

    (2013)
  • K.R. Paap et al.

    Bilingual advantages in executive functioning either do not exist or are restricted to very specific and undetermined circumstances

    Cortex

    (2015)
  • K.R. Paap et al.

    The role of test-retest reliability in measuring individual and group differences in executive functioning

    Journal of Neuroscience Methods

    (2016)
  • S. van der Sluis et al.

    Executive functioning in children, and its relations with reasoning, reading, and arithmetic

    Intelligence

    (2007)
  • N.M. Wöstmann et al.

    Reliability and plasticity of response inhibition and interference control

    Brain and Cognition

    (2013)
  • Antón, E., Carreiras, M., & Duñabeitia, J. A. (2018). On the effects of bilingualism on executive functions and working...
  • E. Bialystok

    Effect of bilingualism and computer video game experience on the Simon task

    Canadian Journal of Experimental Psychology

    (2006)
  • E. Bialystok

    The bilingual adaptation: How minds accommodate experience

    Psychological Bulletin

    (2017)
  • Blackburn, A. M. (2013). A study of the relationship between code switching and the bilingual advantage: Evidence that...
  • H.K. Blumenfeld et al.

    Cognitive control in bilinguals: Advantages in stimulus-stimulus inhibition

    Bilingualism: Language and Cognition

    (2014)
  • C.E. Curtis et al.

    The inhibition of unwanted actions

  • A. de Bruin et al.

    Effects of age on inhibitory control are affected by task-specific features

    Quarterly Journal of Experimental Psychology

    (2018)
  • T. Dijkstra et al.

    The architecture of the bilingual word recognition system: From identification to decision

    Bilingualism: Language and Cognition

    (2002)
  • T. Dijkstra et al.

    Multilink: A computational model for bilingual word recognition and word translation

    Bilingualism: Language and Cognition

    (2018)
  • Donnelly, S., Brooks, P., & Homer, B. (2018). Is there a bilingual advantage on interference-control tasks? A...
  • J. Fan et al.

    Testing the efficiency and independence of attentional networks

    Journal of Cognitive Neuroscience

    (2002)
  • A. Field

    Discovering statistics using IBM SPSS statistics

    (2018)
  • N.P. Friedman et al.

    The relations among inhibition and interference control functions: A latent-variable analysis

    Journal of Experimental Psychology: General

    (2004)
  • T.H. Gollan et al.

    Self- ratings of spoken language dominance: A multilingual naming test (MINT) and preliminary norms for young and aging Spanish-English bilinguals

    Bilingualism: Language and Cognition

    (2011)
  • D.W. Green

    Mental control of the bilingual lexico-semantic system

    Bilingualism: Language and Cognition

    (1998)
  • D.W. Green et al.

    Language control in bilinguals: The adaptive control hypothesis

    Journal of Cognitive Psychology

    (2013)
  • Guido Mendes, C. (2016). The impact of bilingualism on conflict control (Doctoral dissertation). University of Otago,...
  • Cited by (34)

    • Executive functions in mono- and bilingual children: Factor structure and relations with fluid intelligence

      2022, Journal of Experimental Child Psychology
      Citation Excerpt :

      Yow and Li (2015) demonstrated positive associations between age of second language acquisition and interference costs on a Stroop task as well as between balanced bilingualism and performance on a Stroop task and a number–letter task. Moreover, there is still a debate about how the context-relevant language is selected in the bilingual mind, and some researchers have claimed that language selection can happen via facilitation of the target language rather than inhibition of the irrelevant language (Costa et al., 2000; Paap et al., 2019). Thus, inhibitory abilities might not be trained in bilinguals, resulting in comparable performances on inhibition tasks in mono- and bilinguals.

    • Share the code, not just the data: A case study of the reproducibility of articles published in the Journal of Memory and Language under the open data policy

      2022, Journal of Memory and Language
      Citation Excerpt :

      The editor-in-chief gave us the titles of papers reporting quantitative experiments that were published before (N = 59) and after (N = 59) data sharing was made mandatory in JML in 2019. The list of papers published before the open data policy took effect: Akan, Stanley, and Benjamin (2018), Arnold, Strangmann, Hwang, Zerkle, and Nappa (2018), Chan, Manley, Davis, and Szpunar (2018), Chubala, Surprenant, Neath, and Quinlan (2018), Cunnings and Sturt (2018), de Bruin, Samuel, and Duñabeitia (2018), Deliens, Antoniou, Clin, Ostashchenko, and Kissine (2018), Do and Kaiser (2019), Drummer and Felser (2018), Fisher and Radvansky (2018), Frazier and Clifton (2018), Fritz, Kita, Littlemore, and Krott (2019), Fukumura (2018), Galati, Dale, and Duran (2019), Gathercole, Dunning, Holmes, and Norris (2019), Healey (2018), Hopper and Huber (2018), Hsiao and Nation (2018), Isarida, Isarida, Kubota, Higuma, and Matsuda (2018), James, Fraundorf, Lee, and Watson (2018), Jones and Farrell (2018), Jou, Escamilla, Torres, Ortiz, and Salazar (2018), Karimi, Swaab, and Ferreira (2018), Keung and Staub (2018), Kowialiewski and Majerus (2018), Malejka and Bröder (2019), McKoon and Ratcliff (2018), Miller, Gross, and Unsworth (2019), Miyoshi, Kuwahara, and Kawaguchi (2018), Mohanty and Naveh-Benjamin (2018), Nicenboim and Vasishth (2018), Nooteboom and Quené (2019), Osth, Fox, McKague, Heathcote, and Dennis (2018), Paap, Anders-Jefferson, Mikulinsky, Masuda, and Mason (2019), Sahakyan and Malmberg (2018), Scott and Sera (2018), Seabrooke, Hollins, Kent, Wills, and Mitchell (2019), Seedorff, Oleson, and McMurray (2018), Singh, Gignac, Brydges, and Ecker (2018), Slioussar (2018), Stefanidi, Ellis, and Brewer (2018), Susser and Mulligan (2019), Thalmann, Souza, and Oberauer (2019), Uner and Roediger III (2018), Van Bergen and Bosker (2018), van Heugten, Paquette-Smith, Krieger, and Johnson (2018), van Tiel, Pankratz, and Sun (2019), Vasishth, Mertzen, Jäger, and Gelman (2018), Vaughn and Kendall (2018), Veldre and Andrews (2018a,b), Wang, Otgaar, Howe, Lippe, and Smeets (2018), Wedel, Nelson, and Sharp (2018), Wen and van Heuven (2018), Wilson, Donnelly, Christenfeld, and Wixted (2019), Yim, Osth, Sloutsky, and Dennis (2018), Zawadzka, Simkiss, and Hanczakowski (2018), Zhang and Samuel (2018). The list of papers published after the open data policy took effect: Ahn and Brown-Schmidt (2020), Ambrus et al. (2020), Avetisyan, Lago, and Vasishth (2020), Bangerter, Mayor, and Knutsen (2020), Boyce, Futrell, and Levy (2020), Brainerd, Bialer, and Chang (2020), Brainerd, Nakamura, Chang, and Bialer (2019), Brainerd, Nakamura, and Murtaza (2020), Brandt, Aßfalg, Zaiser, and Bernstein (2020), Brewer, Robey, and Unsworth (2021), Bristol and Rossano (2020), Brothers and Kuperberg (2021), Brysbaert (2019), Bürki, Elbuy, Madec, and Vasishth (2020), Chan, Manley, and Ahn (2020), Chetail (2020), Collins, Milliken, and Jamieson (2020), Corps and Rabagliati (2020), Diéez-Álamo, Glenberg, Diéez, Alonso, and Fernandez (2020), Falandays, Brown-Schmidt, and Toscano (2020), Fellman et al. (2020), Floccia et al. (2020), Fox, Dennis, and Osth (2020), Fujita and Cunnings (2020), Gagné et al. (2020), Garnham, Child, and Hutton (2020), Günther et al. (2020), Günther, Petilli, and Marelli (2020), Hesse and Benz (2020), Hollis (2020), Humphreys, Li, Burt, and Loft (2020), Hwang and Shin (2019), Isarida et al. (2020), Jäger et al. (2020), Johns, Jamieson, Crump, Jones, and Mewhort (2020), Kaula and Henson (2020), Lange, Berry, and Hollins (2019), Lauro, Schwartz, and Francis (2020), Lelonkiewicz, Ktori, and Crepaldi (2020), Liang, Ma, Bai, and Liversedge (2021), Li, Ren, Zheng, and Chen (2020), McKinley and Benjamin (2020), Meteyard and Davies (2020), Nooteboom and Quené (2020), Osth, Shabahang, Mewhort, and Heathcote (2020), Reifegerste, Jarvis, and Felser (2020), Saito, Kachlicka, Sun, and Tierney (2020), Samuel (2020), Schubert, Cohen, and Fischer-Baum (2020), Siegelman et al. (2020), Siew, Yi, and Lee (2021), Skrzypulec and Chuderski (2020), Snefjella, Lana, and Kuperman (2020), Snell and Theeuwes (2020), Tirso and Geraci (2020), Troyer and Kutas (2020), Tsuboi and Francis (2020), Villani, Lugli, Liuzza, Nicoletti, and Borghi (2021), Yang et al. (2020).

    • Differences in cognitive processing? The role of verbal processes and mental effort in bilingual and monolingual children's planning performance

      2022, Journal of Experimental Child Psychology
      Citation Excerpt :

      In bilinguals, both languages are jointly activated (Marian & Spivey, 2003; Thierry & Wu, 2007). Diverging theories exist claiming that either this joint activation requires a mechanism to select between the competing linguistic systems (Kroll & Bialystok, 2013) or, alternatively, that bilinguals might also live with the occasional unintended intrusion from the other language (Paap, Anders-Jefferson, Mikulinsky, Masuda, & Mason, 2019). Bilingual language switching is related to a set of specific brain regions, including the left dorsolateral prefrontal cortex and the anterior cingulate cortex (ACC), among others (Abutalebi & Green, 2008; Sulpizio, Del Maschio, Fedeli, & Abutalebi, 2020).

    • The moderating effect of bilingualism on lifespan cognitive development

      2020, Cognitive Development
      Citation Excerpt :

      This task situation parallels the cognitive conflict that bilinguals deal with when having to select the target language in the face of co-activation of the context-irrelevant language. Although most researchers would agree that cognitive conflict tasks require some sort of selection process, there is currently a debate about the exact nature of the mechanisms underlying performance in these tasks (e.g., Paap, Anders-Jefferson, Mikulinsky, Masuda, & Mason, 2019; Rey-Mermet, Gade, & Oberauer, 2018). The examination of the validity of interference tasks as inhibition measures is beyond the scope of the present study; however, in the light of the aforementioned debate, we opted at using the term resistance to interference to refer to goal-directed selection of the target stimulus/response when there is overlapping activation of conflictive information.

    View all citing articles on Scopus
    View full text