Interpretation
Our predictions were based on a combination of previous data and models pertaining to the effects of bilingualism on the brain in adults, together with well-documented general effects of development on the brain. Specifically, we expected that, on top of basic brain developmental patterns, during development greater cortical thickness and volumes (e.g., due to less steep developmental decreases for cortical thickness, that is, less grey matter loss) should emerge for bilinguals as compared to monolinguals, followed by gradual convergence of these metrics between the two groups. (The paucity of adult bilingual evidence from cortical surface area precluded predictions regarding this metric.) In contrast, greater bilingual (vs. monolingual) subcortical volumes and white matter integrity were only expected at later stages of bilingualism, around the same time that cortical differences cease to be observed.
The analyses revealed that while both bilinguals and monolinguals showed the expected general developmental patterns of brain structures, they also showed differences in their developmental trajectories.
First, consistent with the broader developmental literature, both groups showed the following patterns (see Table
4 and Figs.
2,
3,
4,
5). Both evidenced continuous decreases in cortical thickness throughout childhood/adolescence (Remer et al.
2017; Tamnes et al.
2017). In both, cortical volume and cortical surface area generally showed more nonlinear trajectories, often with early increases or stability followed by decreases (Wierenga et al.
2014; Mills et al.
2016; Tamnes et al.
2017). And in both groups, white matter integrity increased continuously with age (Lebel and Deoni
2018; Lebel et al.
2019). Overall, this demonstrates that the patterns observed here follow expected trajectories of brain development.
Second, and of greater interest here, the two groups also showed developmental
differences, which were moreover strikingly similar across measures and structures (Table
4 and Figs.
2,
3,
4,
5). As expected based on the adult literature (see above), greater cortical thickness and cortical volume (less developmental grey matter loss) were observed for bilinguals as compared to monolinguals across multiple cortical regions, with this deviation becoming apparent around late childhood to early puberty (in most cases about age 10–13 for cortical thickness and 12–14 for cortical volume). Moreover, both for cortical thickness and volume, this pattern continued in most regions; though in some cortical areas the values for bilinguals and monolinguals reconverged in early adulthood, about age 20–21. Additionally, across cortical regions, a very early and unexpected difference between the groups was observed: starting from around age 3, the lowest age in our sample, bilinguals displayed lower cortical thickness and volumes than monolinguals, with these values only converging between the groups from around mid to late childhood to early puberty. However, given the relatively small sample size of the bilingual group at the youngest ages (Table
1), this finding should be treated with caution. Interestingly, though we did not have clear predictions for cortical surface area (due to the dearth of studies examining this metric in bilinguals; see above), the one region showing bilingual/monolingual differences in surface area showed a pattern somewhat similar to that observed for cortical volumes, in particular during early childhood. In sum, the bilinguals consistently showed lower values for cortical metrics than monolinguals until mid to late childhood or early puberty, at which point they generally diverged again in showing larger values (especially for cortical thickness and volume), with this bilingual increase continuing into early adulthood, or in some cases reconverging.
Unlike cortical regions, no differences between bilinguals and monolinguals were observed for subcortical structures or the cerebellum. This null effect, which was consistent with our predictions, adds to the literature in suggesting that subcortical and cerebellar effects may only emerge at later stages in more experienced bilinguals, after the initial cortical effects have started disappearing (Filippi et al.
2011,
2020; Pliatsikas et al.
2014,
2017; Pliatsikas
2020).
With respect to white matter, one tract showed bilingual/monolingual differences, that is, striatal–inferior frontal fibers for FA. Strikingly, this again yielded the same pattern as cortical regions, namely lower FA values from age 3 for bilinguals than monolinguals, with the groups gradually converging, till the bilinguals began to show larger values from about age 16. Overall, this suggests a greater increase for bilinguals than monolinguals in white matter integrity over the course of development.
The pattern of results suggests consistency not only in the developmental trajectories of the different metrics, but also in which parts of the brain show effects. In particular, almost all of the bilingual/monolingual by age interactions were found for frontal or parietal structures. This held across all three cortical metrics (thickness, volume, surface area) and the one implicated white matter tract (which connects the striatum and inferior frontal cortex). Only two regions yielded effects outside these two lobes (one in the temporal lobe, one in the occipital lobe), neither of which were among the strongest interactions observed.
How might the implication of frontal and parietal structures in bilingual/monolingual developmental differences, as well as striatal–inferior frontal structural connectivity, be interpreted? Though interpretation must be made with caution, due to the risk of reverse inference (Poldrack
2006,
2011), this pattern appears to be consistent with the involvement of particular circuits (networks).
First of all, according to certain ‘dual-stream’ views, language and audition (as well as vision) are processed by a ventral stream that is closely related to temporal cortex and underlies aspects of auditory object identification and meaning, together with a dorsal stream that projects from parietal (mainly inferior parietal and nearby cortex) to frontal regions (mainly (pre)motor and IFG pars opercularis) and is primarily involved in sensory–motor integration and articulatory functions (Hickok and Poeppel
2007; Rauschecker and Scott
2009). In the present study, the implication both of inferior parietal regions (supramarginal gyrus and inferior parietal) and motor and related regions (precentral, paracentral, as well as posterior portions of superior frontal and perhaps caudal middle frontal) as well as IFG pars opercularis, thus appears to be consistent with changes in the dorsal stream in the developing bilingual brain. Indeed, the supramarginal gyrus, pars opercularis, and superior frontal gyrus showed particularly reliable differences between developmental trajectories for bilinguals and monolinguals (Figs.
2 and
3). The involvement of the dorsal stream would be consistent with evidence suggesting increased contributions from this stream during at least reading in bilinguals as compared to monolinguals (Parker Jones et al.
2012; Bakhtiari et al.
2014), findings that have been attributed to increased articulatory competition in bilinguals (Parker Jones et al.
2012). A similar argument could also be made for increased competition among the different grammars in bilinguals, particularly given the dependence of grammar on pars opercularis and nearby premotor cortex (Friederici
2011). Nevertheless, bilingual-based developmental changes in the dorsal stream do not seem to full explain the observed patterns, not only because changes in other structures were also observed, but additionally because the supramarginal gyrus showed greater effects than more posterior aspects of inferior parietal cortex, which have generally been the focus in language-related dorsal stream models (Hickok and Poeppel
2007; Rauschecker and Scott
2009).
Second, it is possible that the pattern could (also) be due in part to changes in the procedural memory circuit that appears to underlie language (Ullman
2004,
2020). Both the basal ganglia (in particular the striatum) and frontal cortex (in particular (pre)motor regions and the IFG pars opercularis), as well as parietal regions, play key roles in this circuit, which is implicated in the learning, representation, and use of both first and second language (whereas dorsal stream models focus on first language). Procedural memory may underlie multiple portions of language, including grammar, speech–sound representations, articulation, and more generally aspects of both speech production and speech perception (Ullman
2020; Ullman et al.
2020). Thus, the greater involvement of this system in bilinguals than monolinguals would not be surprising, given the need for all these aspects of language to be supported in two languages in bilinguals, as compared to one in monolinguals. The involvement of the striatal–inferior frontal tract is particularly striking, especially given that it yielded the largest bilingual/monolingual by age interaction (see Fig.
5), that is, the greatest difference in developmental trajectories between bilinguals and monolinguals. Interestingly, given the suggestion that the learning of dorsal stream parieto-frontal circuits may depend importantly on procedural memory (Ullman
2004; Ullman et al.
2020), the findings here may be interpreted as implicating both dorsal stream and procedural memory functions as two sides of the same coin.
Third, the patterns observed here may additionally be explained by the involvement of brain structures underlying executive functions. A large literature in both children and adults has implicated a greater role for executive functions in bilinguals than monolinguals, in particular due to the switching of and the control between languages in bilinguals (Valian
2015; Abutalebi and Green
2016). Indeed, such functions have been tied to superior, middle, and inferior frontal regions, the inferior parietal cortex, the precuneus, and the basal ganglia in adults and/or children (Seeley et al.
2007; Mohades et al.
2014; Shen et al.
2019). All of these structures were implicated in the present study, either directly (in cortical grey matter measures) or indirectly (in white matter measures of connecting fibers). Thus, the observed bilingual/monolingual differences in developmental trajectories may be at least partly explained by group differences in executive functions as well as dorsal stream function and procedural memory.
However, the above circuits and functions might not fully explain the bilingual/monolingual developmental changes observed here. In particular, although the precuneus was found to show different developmental trajectories between the groups in all three cortical measures (thickness, volume, surface area), it is not importantly implicated in two of the three circuits and associated functions above, and is not generally implicated with a primary role in executive functions. So what might (additionally) explain its involvement here? The precuneus has been implicated in various functions, perhaps most notably aspects of visuo-spatial processing and declarative memory (Cavanna and Trimble
2006). Intriguingly, some evidence suggests that the precuneous underlies not only visuo-spatial processing, but also the linguistic processing of spatial relations (Wallentin et al.
2008), though it remains unclear why this function should be more engaged in the bilingual than monolingual developing brain. In contrast, declarative memory is clearly involved in both first and second language (working closely together with procedural memory), in particular for lexical knowledge, but also for various other aspects of language, including grammar and speech-sound representations (Ullman
2020; Ullman et al.
2020). Thus, learning, representing, and processing two (or more) languages may engage declarative memory more than one language does. This view is also consistent with the finding here that the posterior parietal and inferior temporal cortex, both of which play important roles in declarative memory and associated functions (Ullman
2016; Tagarelli et al.
2019), show bilingual/monolingual differences in their developmental trajectories. Nevertheless, it remains unclear why other declarative memory structures, in particular the hippocampus and other medial temporal lobe structures, were not involved here—though interestingly, the hippocampus has been found to be enlarged in adult second language learners (Mårtensson et al.
2012), remains plastic in active adult bilinguals (DeLuca et al.
2019b) and its volume declines at a slower rate in older bilinguals as compared to older monolinguals (Li et al.
2017; Voits et al.
2020).
Overall, the pattern of structures showing different developmental trajectories between the two groups seems largely consistent with prior studies of the bilingual brain. In adults, bilingual/monolingual differences have been found mainly in frontal (and the ACC), temporal, and parietal regions, as well as the basal ganglia (mainly the striatum) and thalamus, and a number of (mainly cortico-cortical) white matter tracts (see Sect. "
Bilingualism and brain structure in adults"). In children, the small number of studies (which, however, focused on specific structures) have mainly implicated frontal and nearby cortex (IFG pars opercularis, middle frontal gyrus, the ACC), temporal cortex (STG), parietal cortex (inferior parietal), the basal ganglia (striatum), and tracts related to frontal cortex. The present study, which examined a large sample size across multiple structures and a range of measures, extends and further specifies our understanding of bilingual/monolingual differences in development. In particular, while it confirms the involvement of IFG pars opercularis, middle frontal gyrus, inferior parietal cortex, and (at least connections with) the basal ganglia in the developing bilingual brain, it suggests that certain other structures may also play differential roles in bilingual and monolingual development, in particular, though not only, in frontal and parietal regions.
Implications, limitations, and conclusion
The study has a number of implications and limitations. The findings suggest that bilinguals’ and monolingual’s brains differ even during development. This clearly extends bilingual/monolingual brain differences from adults to the developing brain. The results suggest that the apparent resilience of the bilingual brain to aging (see introduction) may begin already during development, reinforcing the view that any beneficial effects of bilingualism on brain structure may require long-term experience using the two languages (Perani and Abutalebi
2015).
The finding that these patterns overlap to a fair extent with previously observed bilingual/monolingual brain differences in adults both validates the current findings with respect to the adult literature and suggests that some adult brain differences already emerge during childhood and adolescence. This is not in fact surprising given that the vast majority of studies of the adult bilingual brain examined bilinguals who learned more than one language prior to adulthood (Luk and Pliatsikas
2016; García-Pentón et al.
2016; Hayakawa and Marian
2019; Pliatsikas
2019). Nevertheless, the findings also indicate that during development, bilingual/monolingual differences are not exactly the same as those observed in adulthood. In particular, the results suggest that some bilingual/monolingual brain differences found during development disappear by early adulthood, while others appear soon thereafter. Indeed, we saw here that certain bilingual/monolingual differences that emerged by late childhood to early puberty (in particular, greater cortical thickness or volume for bilinguals in a number of regions, likely due to less grey matter loss during development) stabilized, reduced, or even disappeared by early adulthood. Additionally, certain findings that appear to be consistent in the adult literature in more experienced bilinguals were not found or were sparsely observed in the present study, in particular greater bilingual than monolingual subcortical and cerebellar volumes (not observed here) and greater white matter integrity (found in one tract).
Overall, these patterns seem in line with our predictions, namely that developing bilinguals without extensive experience should show greater cortical thickness and volume as compared to monolinguals, though these differences should then gradually decrease or disappear around the same time as they begin to show increased volumes for subcortical structures and greater white matter integrity. Interestingly, the lack of any bilingual/monolingual differences in subcortical volumes and the cerebellum, and the involvement of only one white matter tract, seems to be consistent with the finding that most cortical regions did not show full reconvergence between the groups by the oldest age in the sample, namely 21. These findings thus appear to jibe both with the broad adult literature (regarding both which structures are implicated and their changes over the course of bilingualism) and with our predictions for bilingual/monolingual development differences based in part on this literature.
Nevertheless, a number of limitations suggest the need for further studies. First, we did not predict the finding that at very early ages the bilinguals appeared to show lower values than monolinguals, indeed for nearly all metrics for all structures. It is not clear what may account for this pattern, though the relatively small sample size of bilinguals at these ages suggests treating this finding with caution. Second, the PING database did not contain detailed information regarding the ages of acquisition of bilinguals’ languages, nor the amount or type of language experience they had. The very early bilingual/monolingual effects suggest that the children at this age were exposed to two (or more) languages from an early point, indicating that at least a portion of the sample experienced early ages of acquisition. This also suggests the possibility—though this is speculative—that the rest of the sample might also have experienced quite early exposures to their languages, given that the older children and adolescents may have been drawn from similar subject pools at the same testing sites. This possibility is supported by the observed trajectories of different bilingual/monolingual patterns at different ages, which (as we have argued above) appear to be consistent with experience-based changes over time with similar ages of acquisition across the participants. Nevertheless, the age of exposure of the older children in this sample was not known, and thus future research should examine whether the findings here replicate with populations with clearly documented early ages of exposure, as well as investigating developmental trajectories with later exposures. Third, it may be argued that further measures against type I errors could have been taken. Indeed, we did not correct for multiple comparisons. This decision was taken due to the fact that our approach attempted to balance the likelihood of type I and type II errors, and was already conservative in that it was designed to reduce type I errors by virtue of an analytical procedure akin to a vibration of effects approach, with only those effects with significant results across all alternative analyses being reported. Additionally, the high degree of consistency in the observed patterns in the relative trajectories of bilingual and monolingual brain measures suggests that the reported findings are not spurious. Together, this suggests that type I errors are not prevalent in the findings reported here. Finally, we emphasize that the present study (like the vast majority of developmental studies across large age ranges) is cross-sectional, so caution is warranted in extrapolating to actual developmental patterns within subjects. Nevertheless, we suggest that our findings can provide a solid foundation for well-designed longitudinal developmental studies.
In sum, the present study suggests that bilinguals and monolinguals differ in quite consistent ways regarding the developmental trajectories of brain structures from early childhood to early adulthood. Thus, the study clearly extends previous observations of bilingual/monolingual brain differences in adults to development. The evidence presented here suggests that, as compared to monolinguals, bilinguals show more grey matter (less developmental loss) starting around late childhood and adolescence, mainly in frontal and parietal regions, as well as increased white matter integrity (greater developmental increase) starting in mid-late adolescence, specifically in fibers connecting the striatum and inferior frontal cortex. The findings not only suggest that there may be a developmental basis for some of the structural brain differences found between bilingual and monolingual adults, but also indicate that some bilingual/monolingual differences may occur in the developing but not the adult brain. Overall, the data indicate that the bilingual brain does indeed differ from the monolingual brain, and that this difference begins to be apparent even during development.