Musical training, bilingualism, and executive function: working memory and inhibitory control

Musical training, working memory, and inhibitory control

A body of evidence demonstrates that musicians outperform non-musicians on various cognitive abilities far beyond what the skill involves, such as verbal memory and visuospatial abilities (Costa-Giomi, 1999; Franklin et al., 2008). Beyond specific abilities, musicians have also demonstrated an advantage on broad cognitive abilities such as executive functions. Research evidence supports a strong association between enhanced working memory performance in musicians compared to non-musicians. Improved working memory was found across age groups, including children (Degé, Kubicek, & Schwarzer, 2011; c.f. Fujioka, Ross, Kakigi, Pantev, & Trainor, 2006; Lee, Lu, & Ko, 2007; Moreno et al., 2011; Roden, Kreutz, & Bongard, 2012; Schellenberg, 2011), young adults (Franklin et al., 2008; Meinz & Hambrick, 2010; c.f. Nutley, Darki, & Klingberg, 2014; Oechslin, Van de Ville, Lazeyras, Hauert, & James, 2013; Parbery-Clark, Skoe, Lam, & Kraus, 2009; Strait, Kraus, Parbery-Clark, & Ashley, 2010; Wallentin, Nielsen, Friis-Olivarius, Vuust, & Vuust, 2010; Zuk, Benjamin, Kenyon, & Gaab, 2014), and older adults (Amer, Kalender, Hasher, Trehub, & Wong, 2013; Bugos, Perlstein, McCrae, Brophy, & Bedenbaugh, 2007; Hanna-Pladdy & Gajewski, 2012; Hanna-Pladdy & MacKay, 2011).

Improvements have been found across different measures of working memory, such as simple and complex tasks (the former of which measure memory span or storage, and the latter which recruit additional executive control; Unsworth & Engle, 2006), plus for auditory and visuospatial domains, although there is some inconsistency in findings (Hanna-Pladdy & MacKay, 2011; Hansen, Wallentin, & Vuust, 2013; Lee et al., 2007; Strait et al., 2010). There is also evidence of better inhibitory control in musicians across age groups – in children (Degé et al., 2011; Moreno et al., 2011), young adults (Bialystok & Depape, 2009; Hurwitz, Wolff, Bortnick, & Kokas, 1975; Strait et al., 2010; Travis, Harung, & Lagrosen, 2011), and older adults (Seinfeld, Figueroa, Ortiz-Gil, & Sanchez-Vives, 2013). However, it is not consistently replicated across tasks or studies (Carey et al., 2015; Slevc, Davey, Buschkuehl, & Jaeggi, 2016; Zuk et al., 2014).

Bilingualism, working memory, and inhibitory control

As with musical training, speaking a second language is associated with cognitive improvements outside the skill domain itself. Bilinguals outperform their monolingual counterparts in areas such as metalinguistic abilities and awareness of the mental states of others (Goetz, 2003; Gold, Kim, Johnson, Kryscio, & Smith, 2013). On general cognitive ability, early evidence suggested that bilinguals sometimes outperform monolinguals on non-linguistic interference paradigms, such as the Simon, Stroop, and flanker tasks (Bialystok, Craik, & Luk, 2008; Meuter & Allport, 1999). However, recent investigations using multiple tasks have failed to replicate the bilingual advantage, challenging the consistency of a bilingual cognitive advantage (Kousaie & Phillips, 2012a; Kousaie & Phillips, 2012b; Paap & Greenberg, 2013; Prior, 2012; von Bastian, Souza, & Gade, 2016).

Similarly, although some evidence supports a bilingual advantage in working memory (Bialystok, 2009; Bialystok & Feng, 2009; Luo, Craik, Moreno, & Bialystok, 2013), other studies have challenged this finding (Adesope, Lavin, Thompson, & Ungerleider, 2010; Bonifacci, Giombini, Bellocchi, & Contento, 2011; Feng, Diamond, & Bialystok, 2007; Ratiu & Azuma, 2015; von Bastian et al., 2016). It has been suggested that language effects exist for certain experimental conditions and populations but not others (Bialystok et al., 2017), but the finding of a bilingual effect has not consistently been replicated in other age groups (De Bruin, Treccani, & Della Sala, 2015; Gold et al., 2013, Exp. 2; Kousaie & Phillips, 2012a; Salvatierra & Rosselli, 2010). The overall findings from the bilingualism literature typically demonstrate only a few studies with large interference effects and a bilingual advantage and a majority of studies with smaller effects and little or no bilingual advantage (Duñabeitia et al., 2014). Reviews of bilingual evidence indicate that 80% of the literature yields null effects, with a greater percentage of null results for sample sizes larger than 40 (Paap, Johnson, & Sawi, 2015), in contrast with music where the evidence indicates a small to medium effect size (Sala & Gobet, 2017).

Theoretical mechanisms for benefits from musical training and bilingualism

Independent demonstrations of cognitive associations with the domains of music and bilingualism open up the interesting research question of overlapping mechanisms between these domains that may produce cognitive benefits. The central metaphor is that both skills involve coordination of multiple abilities and domains of performance, and the regular exposure to a context where higher-level cognitive function is constantly in demand may contribute to advanced cognitive performance in experts. Higher-level cognition is theorized to be required for both skills, such as working memory to carry out maintaining and manipulating semantic information (e.g. notes, mechanics, rhythms in music; words, syntax, and timing in language) and inhibitory control in blocking or ignoring competing information internally or from the environment (e.g. external noise, irrelevant notes or words). It is plausible that expert musicians or language speakers, both of whom typically have spent at least ten years practicing their skills, demonstrate enhanced cognition compared to individuals without such training. The need for cognitive functions has been proposed via the executive function hypothesis in musicians (Hannon & Trainor, 2007) and the inhibitory control model in bilinguals (Bialystok, 2009), which later gave way to a more general global cognitive model (Bialystok, 2011).

Theoretical similarities in the processing resources for music and language have been proposed, since both are primarily auditory domains and overlap in organization of syntax and temporal structure (Patel, 2011, 2014; Slevc, 2012; Slevc & Okada, 2015). Indication of shared resources is seen in some bidirectionality between both skill domains; however, transfer from music to language abilities tends to be stronger and more widely supported than in the other direction (c.f. Asaridou & McQueen, 2013; Besson, Chobert, & Marie, 2011). Modularity, that is, separate processing in music, is also supported by evidence from musically impaired individuals, speech perception, and neuroscientific findings (Johansson, 2008; Mok & Zuo, 2012; Peretz, 2006). While both skill forms are cognitively demanding, they have different experience-specific demands (Hutka, Bidelman, & Moreno, 2013; Rogalsky, Rong, Saberi, & Hickok, 2011).

Theories on generalizability of learning or transfer of training support that effects of a given skill depend on the functional demands of learning that skill (Barnett & Ceci, Hickok, 2002; Morris, Bransford, & Franks, 1977). The closer a match between the content and context of a skill domain and a tested ability, the greater the effect observed. The role of similarity in determining cognitive benefits has been supported from evidence on cognition-perception couplings in music, linking aspects of the trained and transfer domains (Carey et al., 2015; Parbery-Clark et al., 2009).

Evidence from direct comparisons between language and music on cognitive abilities indicate that there are domain-specific effects that go beyond the shared processing between domains. Differential effects of each are seen on inhibitory control. Although musicians and bilinguals outperform monolingual musicians on some conflict resolution tasks, musicians showed better performance across tasks (Bialystok & DePape, 2009; Moreno, Wodniecka, Tays, Alain, & Bialystok, 2014). Similarly, music and language conferred benefits for some working memory tasks, but musicians had better performance across tasks (Bidelman, Hutka, & Moreno, 2013; Fiveash & Pammer, 2014).

Current study

In light of recent evidence demonstrating training-induced changes in cognition, the present study investigated the degree to which musical training and bilingualism are associated with enhancements in working memory and inhibitory control. Overall, the current evidence indicates a strong association between musical training and working memory, a possible association with inhibitory control, and limited evidence for a bilingual benefit on either ability. However, several methodological issues preclude causal conclusions from the existing evidence, leaving two main questions: first, which abilities are influenced (if any); and, second, whether music or bilingualism uniquely produce benefits.

Different studies have used different tasks for a given construct, which has led to disparities in findings. Comprehensive investigations using multiple measures per construct have also not shown a consistent bilingual effect across tasks, suggesting that any previously demonstrated benefits may have been task-specific (Duñabeitia et al., 2014; Kousaie & Phillips, 2012a; Paap & Greenberg, 2013). To address this issue and resolve uncertainty in the literature, the present study included multiple tasks that measure each of our constructs of interest (working memory and inhibitory control). A coherent advantage would be demonstrated if there were benefits across several measures of the same construct, rather than a benefit limited to one specific measure.

A second issue arises since the majority of the existing evidence consists of non-experimental investigations comparing groups of skilled experts to non-experts. However, several non-musical and non-language background factors also exist between these groups and could be responsible for some of the observed effects. The largest confounding factors are socioeconomic status (SES) in the bilingual literature and intelligence in the music literature (Corrigall, Schellenberg, & Misura, 2013; Degé et al., 2011; Schellenberg, 2011). Since SES and intelligence are correlated with both our independent and dependent measures, it raises the issue of reverse causality in demonstrated effects (Valian, 2015). Such confounding factors can produce false advantages, or even hide true effects, yet are rarely accounted for in previous investigations (Schellenberg, 2016; Swaminathan & Schellenberg, 2014). To isolate unique effects of musical training and bilingualism, the current study used a sample that was homogenous in educational background and geographical location; and we tested for differences in possible third factors that are known to impact cognitive performance. Most studies have failed to find an effect of bilingualism above that of background variables (Antón et al., 2014; c.f. Calvo & Bialystok, 2014; Mindt et al., 2008; Morton & Harper, 2007; Ratiu & Azuma, 2015), although the evidence from musical training is more favorable (Degé et al., 2011).

To parse out effects unique to musical training, experimental trials have been conducted that randomly allocate untrained individuals who are matched on background factors to either musical training or a different group. Evidence from these trials shows little to no evidence of cognitive benefits of musical training on working memory or interference control (Mehr, Schachner, Katz, & Spelke, 2013; Moreno et al., 2011; Sala & Gobet, 2017). While experimental designs indicate that the small range of benefits observed can be attributed to musical training, they use a short period of training (several weeks to a year) and often an artificial setting. Thus, they address the reverse causality issue but are not comparable to correlational evidence which defines musicians as individuals who have received several years of training and use real-world musical training. Longitudinal designs over several years would address this issue but are practically difficult. More feasible would be correlational studies that control for prognostic variables that predict cognition (non-verbal intelligence, vocabulary, and SES) plus training factors to isolate training-induced effects. Hence, our study compares individuals with a long period of real-life training, while accounting for background and training factors (comparing between those with musical experience and bilingual experience) to address the specificity issue.

There were several predictions for the current study. First, we expected musicians to outperform non-musicians on working memory ability and inhibitory control. In terms of working memory, we expected musicians to have better accuracy than non-musicians on both executive and phonological working memory tasks, because musical training involves manipulation and maintenance of multiple symbols and codes. In terms of inhibitory control, we expected musicians to have faster response times on tasks that require suppression of interfering stimuli, as well as on tasks that require stopping an automatic response, because musical training involves ignoring irrelevant symbols and sounds as well as suddenly pausing and restarting performance when required, particularly in group musical performance.

Second, we expected bilinguals would outperform monolinguals on working memory ability and inhibitory control. As for musical training, it can be argued that bilingualism involves mentally manipulating different symbolic systems. However, the lack of replication in recent studies suggests that there is no coherent evidence for a bilingual inhibitory advantage. Newer explanations propose that a bilingual advantage may be either restricted to certain task conditions or a result of non-language background factors. Our study is especially relevant in light of recent work questioning the existence of a bilingual advantage.

Finally, we were interested in examining whether there would be interactive effects of each skill type and we explored the possibility that being both musically trained and bilingual might confer additive benefits compared to having only one skill type. Given that musical training and bilingualism display some separable effects on cognition, it is possible these effects may interact. One option is that the benefits from each domain may combine to produce an additive effect. Alternatively, each domain may produce a maximal effect, such that no further benefits are possible. Evidence on word learning supports the latter – music and speaking a tonal language have independent and non-cumulative effects (Cooper & Wang, 2012).

Participants

Sample size was determined based on an a priori power analysis and was designed to have 80% power to detect both predicted main effects and the predicted interaction, based on effect sizes of studies reported in prior literature in the target populations of interest. G*Power (Faul, Erdfelder, Buchner, & Lang, 2009) was used to determine sample size. According to the software, a total sample size of n = 128 would be required (i.e. n = 32, with four groups) to obtain a medium to large effect size of Cohen’s f = 0.25 (Cohen, 1988), based on then-published literature, for an analysis of variance (ANOVA; fixed effects, special, main effects and interactions, α-level: 0.05, power [1-β error probability]: 0.80). In this study, we exceeded the necessary sample size. However, a more conservative effect size might have been preferable given null evidence that has emerged in the bilingualism literature since we ran the study.

Bilinguals were fluent in English plus at least one other language (Arabic, Armenian, Bulgarian, Cantonese, Farsi, French, German, Ghanian, Greek, Gujarati, Hebrew, Hindi, Italian, Indonesian, Japanese, Kachi, Korean, Mandarin, Portugese, Punjabi, Romanian, Russian, Serbian, Sinhalese, Spanish, Tibetan, Turkmen, Twi, Urdu, Vietnamese, Yoruba, Zulu). Bilinguals were asked to describe themselves on level of bilingualism on a 5-point scale (i.e. 1 = speak only one language; 2 = weak bilingual; 3 = unbalanced bilingual; 4 = practical bilingual; and 5 = fluent bilingual). All bilinguals were able to speak and understand both languages; and 93% were either practical bilinguals (i.e. can carry out conversation fluently but do not use second language daily) or fluent bilinguals (i.e. able to converse fluently and actively use two languages every day). Crucial factors in defining bilingualism are fluency and proficiency (Luk & Bialystok, 2013). On both of these factors, the level of bilingualism was high and matched previous studies (Bialystok, 2009; Paap & Greenberg, 2013). Individuals were considered to be monolingual if they were fluent only in the English language, with little or no training in a second language. Other language factors that were previously shown to have no effect (such as multilingualism or age of onset; Pelham & Abrams, 2014) were not measured.

Musicians included individuals who had at least eight years of experience playing and performing music, began training at age seven years on average, and regularly practiced music (an average of 6–9 h a week). Musicians had, on average, 12 years of formal musical training, using the Royal Conservatory of Music curriculum or similar, and length of training was in the range of 7–22 years. Moreover, 90% had music theory training, 83% had ear training, and on average musicians rated themselves 3.25 or having “good” sight-reading ability on a 5-point scale where 1 = “beginner” and 5 = “expert.” The majority of musicians in the sample (96%) began their training before the age of 12 years. Musicians consisted of instrumentalists (88%) who played at least 1 of 17 instruments (bass, cello, clarinet, drums, flute, guitar, keyboard, organ, piano, saxophone, shamisen, steel drum, trombone, trumpet, ukulele, viola, violin) and vocalists (12%). Non-musicians included individuals with little (< 2 years) or no exposure to musical training; none currently practiced or performed music.

Measures

Procedure

The working memory and inhibitory control paradigms were a part of a larger battery consisting of 11 tasks that measured various executive constructs. The task battery took approximately 2 h to complete and the tasks were presented to participants in a fixed order, beginning with informed consent, followed by the background questionnaires, the cognitive paradigms, and lastly, debriefing of participants. Cognitive paradigms were the digit span, reading span, stop signal, and Stroop tasks, plus those reported in an earlier publication¹ (Moradzadeh, Blumenthal, & Wiseheart, 2014). The computer-based tasks were presented on a Microsoft Windows XP computer and displayed on a 15-inch (1280 × 1024 pixels) monitor.

Two extra tasks were added to the battery to improve the initial choice of tasks. First, the flanker task was added to the battery one-third of the way through initial data collection and data were collected for 95 participants on this task (Appendix 1). After testing had been completed, we realized that one of the tasks (the reading span) may have unfairly disadvantaged the bilingual group due to the task’s emphasis on English language ability. Hence, a non-verbal task (the operation span) was added to the battery to clarify any issues that bilinguals may have had on the reading span task and assess whether data from this task corroborated the initial findings. The initial participants were re-contacted one year after commencing data collection to see if they would be willing to participate in an additional task. Of the 153 participants in the original study, 54 participants could be reached and agreed to partake in the follow-up session. Of this subset, all participants had partaken in the first testing session and demographic variables for this group were consistent with the initial 153 participants.

Analyses

Between-subjects 2 × 2 (musician status [musician, non-musician] × language status [bilingual, monolingual]) ANOVAs and Bayesian analyses were conducted for each measure. Data analyses for all tasks were performed using more stringent classification criteria for bilingualism and musician status (Appendix 2 and 3). These secondary analyses produced results consistent with the ones reported in the primary text.

To account for using multiple measures, we used the step-wise Bonferroni-Holm procedure.² Based on guidelines for multiplicity (Cribbie, 2017) the alpha level was adjusted across all the tests for each construct but not across constructs, since we were interested in how our independent variables (musical training or bilingualism) affected each construct separately. Adjusted significance values at an alpha level of 0.05 for three measures per construct were p _adjusted = 0.017, p _adjusted = 0.025, and p _adjusted = 0.05, in order of the size of obtained effects. Results remained significant after accounting for multiplicity.

Background variables

There were no significant differences between groups on age, F(3, 148) = 1.63, p = 0.185, or SES (as measured by mother’s education level), F(3, 149) = 2.24, p = 0.086 (Table 1). Bayesian analyses also showed no notable effect across age for music, language, or an interaction, BF_Inclusion = 0.15, BF_Inclusion = 0.15, BF_Inclusion = 0.16, respectively, nor for SES, BF_Inclusion = 0.43, BF_Inclusion = 0.34, BF_Inclusion = 0.25. Non-verbal IQ did not differ between groups, F(3, 148) = 0.26, p = 0.857, similarly for Bayesian analyses for music, language, or an interaction, BF_Inclusion = 0.35, BF_Inclusion = 0.22, BF_Inclusion = 0.08, respectively. Vocabulary score was significantly higher in musicians compared to non-musicians, t(70) = 2.88, p < 0.005, d = 0.68, BF_Inclusion = 5.64, and higher among monolinguals compared to bilinguals, t(70) = 3.47, p < 0.001, d = 0.82, BF_Inclusion = 25.05, with moderate to strong differences supported by Bayesian analyses.

Working memory

The reading span task showed a main effect of musician status, F(1, 66) = 8.74, p = 0.004, with a small to medium effect size, η² = 0.098, and strong evidence from Bayesian analyses, BF_Inclusion = 18.27. Musicians correctly recalled more items (mean = 37.2, SEM = 2.6) than non-musicians (mean = 25.5, SEM = 2.9), after controlling for the effects of receptive vocabulary (Fig. 1). Analyses of covariance controlling for receptive vocabulary found neither a significant main effect of language status nor a significant musician by language status interaction.

The digit span backward task showed a main effect of musician status, F(1, 148) = 6.07, p = 0.015, and a small effect size, η² = 0.04. Musicians correctly recalled more items (mean = 8.43 items, SEM = 0.274) than non-musicians (mean = 7.44 items, SEM = 0.294; Fig. 2). No other effects reached significance. Bayesian analyses supported an inconclusive but possible music effect, BF_Inclusion = 2.31, and inconclusive to no effects for language or an interaction.

The operation span task showed a main effect of musician status, F(1, 50) = 4.62, p = 0.018, with a medium to large effect size, η² = 0.11, and strong evidence from Bayesian analyses, BF_Inclusion = 11.93. Musicians correctly recalled more items (mean = 44.8, SEM = 3.25) than non-musicians (mean = 36.2 items, SEM = 4.00; Fig. 3). No main effect of language was demonstrated. The analysis showed an interaction between musician and language status, F(1, 50) = 5.44, p = 0.024, with a medium to large effect size η² = 0.16, and strong evidence from Bayesian analyses, BF_Inclusion = 17.85. A follow up t-test demonstrated higher accuracy in monolingual musicians (mean = 52.7, SEM = 4.14) compared to monolingual non-musicians (mean = 26.9, SEM = 4.82) on this task, t(30) = 3.70, p = 0.001. However, no significant difference was found on this task between bilingual musicians and non-musicians (p > 0.940). Bayesian analyses also found moderate evidence for a language effect, BF_Inclusion = 4.04.

Inhibitory control

On the flanker task, a difference score was calculated by subtracting the mean reaction times for congruent trials from incongruent trials. Results showed a main effect of musician status, F(1, 92) = 8.18, p = 0.005, with a small effect size, η² = 0.012, with faster response times (i.e. smaller difference scores) among musicians (mean = 57.4 ms, SEM = 4.78) than non-musicians (mean = 65.4 ms, SEM = 5.76; Fig. 4). Bayesian analyses revealed inconclusive congruency by music interaction, BF_Inclusion = 0.81, demonstrating that even though the p value is significant, the effect magnitude is not meaningful to interpret. No other effects reached significance.

A secondary analysis was run to compare music and language groups on baseline conditions of the flanker task (average of null and neutral) and mixed conditions (average of congruent and incongruent). Musician groups were significantly faster than non-musician groups on baseline conditions, F(1, 91) = 7.06, p = 0.009, η² = 0.07, and on mixed conditions, F(1, 91) = 8.02, p = 0.006, η² = 0.08. No differences were found between language groups on any of the conditions (ps > 0.406). Independent samples t-tests examining performance of musicians and non-musicians on the flanker task revealed significant differences in RT on all conditions, including null (t(94) = − 3.04, p = 0.003), neutral (t(94) = − 2.25, p = 0.027), congruent (t(94) = − 2.57, p = 0.012), and incongruent (t(94) = − 3.10, p = 0.003) trials, suggesting a possible processing speed advantage on musicians across on response times for all types of trials, including null, neutral, congruent, and incongruent conditions. However, since a similar benefit was not seen on the other measures, a speed of processing benefit remains inconclusive.

A third analysis was conducted to compare our findings to previous findings that used an identical version of the task (Bialystok et al., 2017). A proportion score was calculated by subtracting baseline conditions from mixed conditions and dividing by the baseline conditions. No difference was found in either music or language groups on the proportion score (ps > 0.35).

Difference scores on the Stroop task were calculated by subtracting congruent from incongruent conditions. The Stroop task showed a trend towards significance for musician status, F(1, 147) = 3.56, p = 0.061, with a small effect size, η² = 0.023. Bayesian analyses revealed an inconclusive music effect, BF_Inclusion = 0.77, so the marginal p value should not be interpreted. No other effects reached significance.

The stop signal task demonstrated no significant differences between musicians and non-musicians in their performance on no signal trial SSRTs in the stop signal task (p > 0.22). We considered the possibility that some musicians may not have been as proficient in an orchestral or group musical environment where response inhibition is necessary. Thus, in order to analyze differences on response inhibition in the relevant musician group, the performance of musicians who had specifically indicated that they were a part of an orchestra or band was compared to those without musical training on the stop signal task. Results of this follow-up analysis also did not indicate any musician benefits on SSRT, p > 0.76. No other effects reached significance, all ps > 0.90, all BF_Inclusions < 0.33.

No evidence for a bilingual advantage on cognition

In line with emerging evidence suggesting no bilingual advantage on cognitive tasks, our study failed to demonstrate associations between being bilingual and improved performance on working memory or inhibitory control. Our findings address the main two proposed reasons for null effects in the bilingual literature: use of single tasks and a “peak advantage” of bilingual young adults (Bialystok, Martin, & Viswanathan, 2005; Paap & Greenberg, 2013). Both reasons fail to explain the lack of a bilingual advantage in the current study, since multiple tasks were used and young adults did show a benefit on musical training, adding further skepticism to the notion that a general bilingual advantage exists beyond task-specific or sample-specific constraints.

Beyond task effects, another debate is that the bilingual advantage may be restricted to certain individuals. A few previous studies have found a larger bilingual advantage in older than in younger adult participants, but most did not (c.f. Abutalebi, Canini, Della Rosa, Green, & Weekes, 2015; Bialystok et al., 2008; Bialystok, Craik, Klein, & Viswanathan, 2004; Faroqi-Shah, Sampson, Pranger, & Baughman, 2016; Gold et al., 2013, Exp. 2). Notably, a bilingual effect in older adults has not been consistently replicated, while benefits of musical training have been found across age groups.

One explanation for null effects in young adult bilinguals is that young adults are already at maximal executive function performance (Bialystok et al., 2005). The “functional ceiling” was proposed due to the number of opportunities available for individuals to reach peak cognitive performance, including daily demands such as inhibiting distractions and unsuitable responses (Bialystok, Craik, & Luk, 2012). However, there is no research demonstrating that the hypothesized asymptotic limit exists for executive function performance, and a direct test of the ceiling hypothesis found within experiment improvement (Paap, Johnson, & Sawi, 2014; Paap et al., 2014, May), thus challenging the hypothesis. Our demonstrated musical training benefits, even in young adults, support similar musician but not bilingual benefits in our previous study with undergraduates (Moradzadeh et al., 2014) and suggests that an age limit explanation does not hold.

It has been argued that the task selection may hide a bilingual advantage (Bialystok, 2009). One factor is complexity of the task, with the bilingual advantage being strongest with complex tasks requiring substantial amounts of executive function (Bialystok, 2009, 2011). Simple working memory and short-term memory tasks showed no difference in bilinguals and monolinguals (Bialystok & Feng, 2009). However, others found the opposite, where a bilingual advantage was seen on a simple but not complex version of the Simon task (Salvatierra & Rosselli, 2010). Granted, these are different tasks, but the set of findings still challenges the hypothesis of a general benefit in bilinguals. Despite using simple and complex forms, no bilingual benefits were found in our study. In comparison, musical training demonstrated improvements on both simple and complex working memory measures, consistent with previous data indicating the same (Lee et al., 2007). Another proposition is that the bilingual advantage can only be observed under certain experimental conditions—for example, on high-monitoring conditions in the flanker task, or in baseline compared to mixed conditions (Bialystok et al., 2017; Costa, Hernández, Costa-Faidella, & Sebastián-Gallés, 2009). However, we used the same tasks and analysis procedure as previous investigations and could not find a bilingualism advantage.

Another factor is verbal content of tasks, which could even reverse a bilingual advantage (Bialystok, 2009; Bialystok, Poarch, Luo, & Craik, 2014). Correlations of vocabulary with reading span scores in our study echo concerns that bilinguals may be disadvantaged on verbal measures. However, even after following recommendations to covary vocabulary level (Barac, Bialystok, Castro, & Sanchez, 2014), no change in results was seen, which suggests that possible verbal deficits in bilinguals did not affect performance on the reading span task. Further, we did not find an effect on complex non-verbal tasks—for both the operation span task, a non-verbal measure of working memory, or the digit span backward, an auditory measure. Our null effects match those of Ratiu and Azuma (2015) who also did not find a bilingual benefit when using complex non-verbal tasks. Even where an effect was previously found on a non-verbal task, it was not consistent. For example, bilinguals performed significantly better on a backward version of the Corsi blocks task, but not on the forward version, or on the self-ordered pointing task, a complex non-verbal measure (Bialystok et al., 2008). Overall, beyond task-specific effects (verbal content or complexity), for a coherent benefit to occur it would need to be demonstrated across measures of a construct.

While previously proposed conditions for a bilingual advantage were not supported by our investigation, it is possible that our study may have missed a factor that is required for a bilingual advantage. We concur that a bilingual advantage may be generated under very specific conditions, but the very large number of suggested conditions and requirements shows that any effect is the exception rather than the norm and may be a statistical artifact. Nevertheless, learning a language has intrinsic benefits regardless of whether cognitive benefits exist.

Methodological considerations

Several caveats should be noted in relation to the current findings. First, the correlational design of the present study prevents us from knowing whether musicians had pre-existing differences from non-musicians before undergoing musical training. Although multiple tasks per measure were used, due to the unbalanced sample sizes across tasks we could not run a factor analysis or multivariate ANOVA. An analysis would have been ideal to account for individual variability between groups and any co-linearity between variables, but would have produced a considerable loss in power as it would need to use the smallest possible sample size across task. However, a multivariate analysis was conducted by combining data across tasks to remove individual differences and found no change in the pattern of results. Future research could address this with larger sample sizes and multivariate analyses.

The current findings are limited to the classifications of musical training and bilingualism used. Musicians were defined as individuals who had formal musical training of at least eight years (the final sample had an average of 12 years of training). The amount and type of training matches previous studies (Fujioka et al., 2004; Lee et al., 2007; Schellenberg, 2011), with a minimum number of years being important since extra-musical effects are linked to the duration of training (Wan & Schlaug, 2010). Since effects of musical training on cognition were found to not vary for instrumental and vocal training (Bialystok & DePape, 2009), we did not distinguish between these. Similarly, our study based the definition of bilinguals (proficiency and frequency of usage) on empirically supported components of language use (Luk & Bialystok, 2013). Other language factors that were previously shown to have no effect (such as multilingualism or age of onset; Pelham & Abrams, 2014; Vega-Mendoza, West, Sorace, & Bak, 2015) were not measured.

The classification criteria we used for musicians and bilinguals were selected to make both groups qualitatively similar. Rather than quantify both groups on the same measurements of experience for each skill, we chose to use the strongest available definitions in the independent bilingualism and music literature to classify each, as this provides reliable definitions that also connect to previous findings. Although the definitions used mean that the measure of experience for skill form is not directly matched, they enabled us to measure individuals who had high levels of ability on their respective skills forms (either language or musical training), rather than being comparable only on quantitative information, but different in their level of ability. Hence, we do not think that similar quantitative information is comparable. For example, four years of musical training is qualitatively different to four years of bilingual experience. Comparable quantitative details on how both types of experience relate to other participant characteristics are presented in Table 1. Future research can investigate how within-group variability in training influences results.