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

Abstract

Three studies compared bilinguals to monolinguals on 15 indicators of executive processing (EP). Most of the indicators compare a neutral or congruent baseline to a condition that should require EP. For each of the measures there was no main effect of group and a highly significant main effect of condition. The critical marker for a bilingual advantage, the Group × Condition interaction, was significant for only one indicator, but in a pattern indicative of a bilingual disadvantage. Tasks include antisaccade (Study 1), Simon (Studies 1–3), flanker (Study 3), and color-shape switching (Studies 1–3). The two groups performed identically on the Raven’s Advanced Matrices test (Study 3). Analyses on the combined data selecting subsets that are precisely matched on parent’s educational level or that include only highly fluent bilinguals reveal exactly the same pattern of results. A problem reconfirmed by the present study is that effects assumed to be indicators of a specific executive process in one task (e.g., inhibitory control in the flanker task) frequently do not predict individual differences in that same indicator on a related task (e.g., inhibitory control in the Simon task). The absence of consistent cross-task correlations undermines the interpretation that these are valid indicators of domain-general abilities. In a final discussion the underlying rationale for hypothesizing bilingual advantages in executive processing based on the special linguistic demands placed on bilinguals is interrogated.

Introduction

Fluent bilinguals have extensive experience in language switching that involves monitoring the situation to select the appropriate language, activating the selected language, and inhibiting the other language. This extensive practice may lead to an enhanced ability in cognitive control that is general and not language specific. Indeed several investigators have reported a bilingual advantage in tasks that seem to require executive processing (EP), that is, the ability to monitor goal-setting cues, to switch attention to goal-relevant sources of information, and to inhibit those that are irrelevant or competing (Bialystok, 2006, Bialystok et al., 2004; Bialystok, Craik, & Luk, 2008; Costa, Hernández, & Sebastián-Gallés, 2008). However, as Hilchey and Klein (2011) enumerate in their recent comprehensive review, there have also been many failures to observe a bilingual advantage.

Although the construct of EP continues to evolve, it is often viewed as a set of interrelated component processes all involving the prefrontal cortex (PFC) with each component recruiting other constellations of cortical function. This componential framework allows for the possibility that the related components have some degree of anatomical and functional independence. Ideally an investigation of bilingual advantages in EP should be grounded in a specific conceptual framework that elucidates the nature of executive processes and guides operational definitions for manipulating and measuring them.

As discussed in Section 1.1.1 there is very little evidence that the measures and tasks typically used to test for differences between bilinguals and monolinguals in inhibitory control are tapping into the same general ability. This is somewhat surprising because the seminal work of Miyake and Friedman (Miyake et al., 2000, Friedman et al., 2008, Miyake and Friedman, 2012) shows evidence for three components of EP: updating, switching¹, and inhibition. In their large-scale studies these investigators conducted confirmatory factor analyses (CFA) using measures from three different tasks for each of the three hypothesized components. For each latent variable (viz., updating, switching, inhibition) the three observed variables (e.g., color, number, category) linked to the same latent variable (e.g., switching), are correlated with one another, and result in standardized factor loading ranging from .40 to .71. However, it is instructive to note that three of the four lowest factor loadings are for the inhibition tasks. At the higher level the three latent variables (updating, switching, inhibition) correlate with one another and this is consistent with the assumption that each contributes to a common EP. When the same data are reanalyzed with a second order CFA where the three latent variables (updating, switching and inhibition) are nested under a common EP latent variable, the nine observed measures all load on common EP with two of the nested components (updating and shifting) still making unique contributions. In summary, these studies support the theory of a general EP ability with separable updating and switching components and an inhibition component that is not separable and that is weakly linked to the general EP ability.

Most tests for a bilingual advantage in EP in adults have focused on only two of three components studied by Miyake and Friedman (viz., switching and inhibition) and frequently employ tasks not tested by Miyake and Friedman. As discussed in the next section this different mix of specific tasks has resulted in less convergent validity.

In Bialystok et al.’s seminal article primary focus was placed on the proposition that bilinguals are better at selecting goal-relevant information and suppressing competing and distracting information. Bilinguals exercise this type of control at two levels: (1) at a high level of goal setting when one language is selected and the other is inhibited and (2) at a lower level where the lexical forms of the goal relevant language are activated and the competing translation equivalents are inhibited (e.g., Green, 1998). If this extensive practice hones a general ability, not specific to language, then bilinguals should be less vulnerable to interference in nonlinguistic tasks.

The standard marker of inhibitory control is the difference in mean response time between trials that require conflict resolution compared to those that do not. In the Stroop, Simon, and the Eriksen flanker tasks conflict occurs on a subset of trials because a potent but task-irrelevant stimulus is often paired in an incongruent manner with the task-relevant stimulus. Performance can be enhanced by boosting the influence of the goal-relevant information relative to that of the competing information. The effectiveness of this control can be inferred from differences in response time between congruent trials and incongruent trials with smaller interference effects implying superior ability. In a recent and comprehensive review of bilingual advantages in EP Hilchey and Klein (2011) conclude that evidence for a bilingual advantage in inhibitory control is rare in both children and young adults.

In the present studies inhibitory control should play a role in the Simon task (Studies 1–3), antisaccade task (Study 1), and Eriksen flanker task (Study 3). Thus, the critical test for a bilingual advantage in inhibitory control is the presence of a significant Group (bilingual versus monolingual) × Trial Type (congruent versus incongruent) interaction with the pattern of interaction showing a larger interference effect for the monolingual group. Compelling evidence for a bilingual advantage in inhibitory control would demonstrate significant advantages in two or more tasks requiring inhibitory control and further show that the interference effects correlate with each other as one would expect if each task includes a common component associated with a general ability to exercise inhibitory control.

Although researchers investigating the bilingual advantage have employed several different tasks that should require inhibitory control the same set of matched bilinguals and monolinguals typically participate in only a single task and, hence, provide only a single measure of this component. This is less than satisfactory because the three most-frequently used nonlinguistic interference tasks do not correlate with one another. Stins, Polderman, Boomsma, and de Geus (2005) tested a group of 12-year old children using the flanker, Simon, and Stroop interference tasks. The correlations between these tasks were all smaller than +0.20 and nonsignificant. Fan, Flombaum, McCandliss, Thomas, and Posner (2003) reported that the flanker, Simon, and Stroop tasks all activated the AC and the left PFC, but again the interference scores were uncorrelated. Each task also activated areas unique to that task. They conclude that “The behavioral and fMRI results taken together seem to argue against a single unified network for processing conflict, but instead support either distinct networks for each conflict task or a single network that monitors conflict with different sites used to resolve the conflict” p. 42. Likewise, Kousaie and Phillips (2012a) using a sample of 51 young adults report no significant correlations between the Stroop, Simon, and Eriksen flanker tasks. Keye, Wilhelm, Oberauer, and van Ravenzwaaij (2009) report a structural-equation analysis of the data obtained from 150 adults who participated in both the flanker and Simon task. There was no association between the two interference tasks. Likewise, Humphrey and Valian (2012) using a sample of 208 young adults report no significant correlation between the Simon and flanker effect. In contrast to all of the above comparisons involving the Simon, flanker, or Stroop tasks; Unsworth and Spillers (2010) in a study using college students do report a significant correlation (r = +0.17, p < .05) between a flanker and Stroop task. The weak correlation achieves statistical significance because the n of 181 is very large.

The fact that Friedman et al. (2008) found evidence for an inhibition component is at odds with the absence of significant correlations reported above for the Simon, Stroop, and flanker tasks. At first look this may appear puzzling since the nature of the inhibition required in the three tasks used by Friedman et al. (antisaccade, stop signal, and Stroop) appears to be as varied, if not more so, than the type of inhibition required across the Simon, flanker, and Stroop tasks. Although these tasks always produce robust interference effects, individual differences seem to be another matter. Friedman et al. (2008) point out that the individual difference correlations in interference tasks are usually low and seem sensitive to task variations. The present studies will provide additional tests of convergent validity between the Simon and antisaccade tasks (Study 1) and between the Simon and flanker tasks (Study 3).²

Given that cognitive control dynamically changes in response to changing goals and changing affordances another important type of executive process is monitoring one’s performance, internal states, and current environment.³ This possibility was first proposed by Bialystok et al. (2004) in the context of the Simon task: “The advantage for bilinguals, therefore, may be not in the enhanced ability to inhibit the misleading spatial cue but in the ability to manage attention to a complex set of rapidly changing task demands” p. 292. Costa, Hernández, Costa-Faidella, and Sebastián-Gallés (2009) have also focused more on monitoring than inhibitory control: “The bilingual advantage in overall RTs may reveal the better ability of bilinguals to handle tasks that involve mixing trials of different types: bilinguals would be more efficient at going back and forth between trials that require implementing conflict resolution and those that are free of conflict” (p. 136). Costa et al. have also updated the underlying rationale for why managing two languages should enhance conflict monitoring: “This better functioning of the monitoring system may come about because of the bilinguals’ need to continuously monitor the appropriate language for each communicative interaction. That is, proper communication in bilingual settings involves the monitoring of the language to be used depending on the interlocutor(s) language knowledge”. (Costa et al., 2009, p. 136).

Differences in monitoring between bilinguals and monolinguals have been inferred from two different measures. The arguably better indicator is the difference in mean RT between a pure block of easy choice RT trials that involve no conflict and the mean RT on the congruent trials in a block that mixes both congruent and incongruent trials together. This difference should reflect the cost of having to monitor for the presence of conflict, apart from the need to actually resolve the conflict on incongruent trials. Ideally, the two groups would show comparable RTs on the pure blocks of easy trials showing that they are matched in terms of speed when there is no need to monitor for potential conflict. A control condition of no-conflict trials is not always included in the experimental design. Under these circumstances a bilingual advantage in monitoring has been inferred if bilinguals are faster than monolinguals on the congruent trials.

Another special experience of bilinguals is that they get lots of experience in switching per se as they shift from one language to the other. Language switching is complicated by the common assumption that both lexicons are connected to the same conceptual system and that consequently switching leads to the need to inhibit the translation equivalents in the non-target language. Setting this complication aside, the ability to switch from one task to a completely different task is assumed to require an executive process (“switching”) that is functionally separable from inhibitory control and monitoring and involves unique areas in addition to the anterior cingulate (AC) and PFC (Collette et al., 2005).

Tests for bilingual advantages in switching include three critical conditions: (1) pure blocks in which participants perform the same task on every trial, (2) mixed-block trials that involve a switch from one task to the other, and (3) mixed-block trials that repeat the same decision. For example, in a mixed block Prior and MacWhinney (2010) presented a precue on every trial that signaled to the participant whether to make a binary color decision with two fingers of one hand or a binary shape decision with two fingers of the other hand. The difference between repeat trials and switch trials in the mixed block was used as an indicator of “switching costs” whereas the difference between repeat trials and pure trials was used as an indicator of “mixing costs”. The results showed a bilingual advantage in switching costs, but not mixing costs. These results, in isolation, are coherent if switching and monitoring are two separate EP components and bilingualism provides special and domain-independent experience only for switching.

The main purpose of this study is to use the framework and strategy presented above to determine if there is a coherent set of evidence for a bilingual advantage in EP. Three large samples of participants complete multiple tasks that require EP. Each task yields multiple comparisons that are typically assumed to be associated with specific components of EP. More specifically, the analyses of each sample, and the combined sample, enable multiple tests for bilingual advantages in inhibitory control, monitoring, and switching. Furthermore, by examining the correlations between performance variables one can assess the assumption that these markers are converging on domain general measures of cognitive control.