Enhanced efficiency in the bilingual brain through the inter-hemispheric cortico-cerebellar pathway in early second language acquisition

The study investigates how second language (L2) learning, particularly the age of acquisition (AoA), shapes whole-brain functional organization. Drawing on evidence that the brain’s network properties such as segregation (modularity) and integration (global efficiency) relate to cognition, the authors hypothesized that early L2 learning (including simultaneous bilingualism from birth) would enhance global whole-brain efficiency and potentially modularity. Given increasing evidence for cortico-cerebellar involvement in language processing and bilingual cognitive control, the research aims to determine how the cerebellar system interacts with cortical and subcortical regions across different L2 AoA. Using resting-state functional connectivity and whole-brain connectome analyses, the study tests whether bilingualism, and the timing of L2 learning, alter brain network integration and segregation and identifies the network features and pathways driving any observed effects.

Prior work indicates bilingualism can positively affect working memory, cognitive control, attention, speech-in-noise perception, aging-related cognition, and neurological recovery, with outcomes modulated by task, age, and learning environment. Neuroimaging has linked bilingualism to functional and structural brain changes, shifting the focus from localized language regions to distributed networks engaging frontal, temporal, parietal, and cerebellar areas. The cerebellum is implicated in rule processing, syntax, and error correction, with right posterior cerebellum and left frontal regions commonly co-activated during language tasks. Network science has characterized the brain as modular yet integrated, where global efficiency supports cognitive performance, including language. Past resting-state studies in bilinguals primarily examined language and control networks, with findings of enhanced frontal and inter-parietal connectivity and AoA-related inter-hemispheric frontal connectivity. Few studies have evaluated whole-brain network properties across widespread regions including the cerebellum; one prior study with a smaller sample did not detect whole-network differences and excluded cerebellar regions. Collectively, existing evidence suggests that bilingualism influences broad network interactions, and that AoA may be central to these effects.

Sample: Resting-state fMRI data from 151 right-handed, healthy participants were compiled from studies conducted at the Montreal Neurological Institute (MNI) using the same scanner and parameters. Grouping was based on self-reported AoA for French/English: simultaneous bilinguals (both languages from birth), early bilinguals (AoA ≤ 5 years), late bilinguals (AoA > 5 years), and monolinguals (limited L2 exposure). Exclusion criteria included any language, hearing, or visual impairment; traumatic brain injury; medical or neurological disorder; and MRI incompatibility. Ethics approval was obtained and informed consent provided. MRI acquisition: 3T Siemens TrioTim with 32-channel head coil. Resting-state T2*-weighted EPI: 132 volumes, TR/TE = 2260/30 ms, flip angle 90°, matrix 64×64, FoV 224 mm, 38 slices, slice thickness 3.5 mm, duration 5:04 min; eyes open fixation. T1 MPRAGE: TR/TE = 2300/2.98 ms, flip angle 9°, matrix 256×256, FoV 256 mm, slice thickness 1 mm. Preprocessing: CONN v20b (SPM12 routines). Steps included realignment/unwarping, outlier detection (ART; intensity >3 SD or motion >0.5 mm), segmentation, normalization to MNI space, and 6 mm FWHM smoothing. Nuisance regression included aCompCor components (WM/CSF), motion parameters, and outlier volumes. Connectome construction: 374 ROIs (333 cortical per Gordon parcellation, 15 subcortical, 26 cerebellar from Harvard-Oxford). ROIs grouped into 14 modules: 12 cortical functional modules plus subcortical and cerebellar modules. Functional connectivity edges computed as Pearson correlations between ROI-averaged time series and Fisher r-to-z transformed; negative edges discarded; weighted analyses applied. Network features: Global efficiency (E) computed as the average nodal efficiency (inverse shortest path lengths to all other nodes; Dijkstra’s algorithm) across the network. Modularity (Q) estimated using Newman’s algorithm, reflecting the proportion of within-module to all possible connections. Implementation via R igraph. Statistics: Whole-brain network group effects tested by one-way ANOVA with Welch’s t-tests for pairwise comparisons; multiple comparisons controlled via Bonferroni-Holm. AoA effects tested (within early and late bilinguals) using Pearson’s correlation. Network-Based Statistics (NBS) identified clusters of edges contributing to group differences, with an a priori edge threshold p < 0.01 (two-sided) and 1000 permutations, FWE-corrected cluster testing by sum of connectivity strengths; recursive NBS applied to ROI-level connections within significant network clusters. Consistency checked with p < 0.05 threshold. Additional analyses: Linear regressions assessed relationships between years of L2 experience (YoE = age − AoA) and L2 proficiency (self-reported 1–7 across reading, speaking, writing, comprehension) with E, Q, and NBS network connectivity. Musical training (binary; 26 participants) entered as a covariate in whole-brain and network analyses.

- Whole-brain integration: Significant group effect for global efficiency (E) across monolinguals and bilingual subgroups (F3,147 = 3.39; pFWE = 0.04). Early bilinguals and simultaneous bilinguals showed higher E than monolinguals (early: t40.3 = 3.26; pFWE = 0.014; simultaneous: t42.7 = 2.77; pFWE = 0.042), while late bilinguals did not differ significantly (t36.2 = 1.88; pFWE = 0.272). No significant differences in E among the three bilingual groups (ps > 0.074). - Network segregation: No group effects for modularity (Q) (F3,147 = 0.53; p = 0.66). - Age of L2 acquisition: Within bilinguals, AoA correlated negatively with E (rho = −0.22; p-uncorrected = 0.037), indicating earlier AoA associates with higher efficiency; AoA was not significantly related to Q (rho = −0.14; p = 0.18). - NBS network-level connectivity: Compared to monolinguals, simultaneous bilinguals exhibited higher cerebellar connectivity with association networks including default mode, dorsal and ventral attention, fronto-parietal, and sensorimotor-hand networks, and higher ventral attention–salience connectivity (pFWE = 0.001). Early bilinguals showed higher cerebellar connectivity with default mode, dorsal attention, and fronto-parietal networks, and higher auditory connectivity with default mode and ventral attention (pFWE = 0.001). - ROI-level connectivity: Identified 118 stronger edges in simultaneous bilinguals versus monolinguals. A greater proportion were inter-hemispheric vs intra-hemispheric (64:37; chi-square(1) = 7.22; p = 0.0072), with right cerebellar lobules VI and VIII showing numerous inter-hemispheric connections to left dorsal attention areas (ten connections). - Bilingual experience measures: No significant relationships of L2 proficiency or years of L2 experience with E, Q, or NBS-measured functional connectivity. Accounting for music training did not alter the main group differences.

Bilingualism, especially when L2 is acquired early or from birth, is associated with greater whole-brain functional integration without changes in modular segregation. This suggests that the organization of the bilingual brain emphasizes enhanced inter-module interactions rather than altered community structure. Network-based analyses indicate that cortico-cerebellar circuits drive the observed efficiency differences, underscoring the importance of cerebellar interactions with association cortical networks for language function and learning beyond classical perisylvian regions. Simultaneous bilinguals show pronounced inter-hemispheric connectivity, aligning with prior structural and functional findings implicating the corpus callosum and inter-hemispheric pathways in bilingual experience and L2 learning success. AoA emerged as a key factor influencing network organization, whereas L2 proficiency and years of exposure did not show associations, pointing to the timing of language exposure as a dominant determinant of functional network efficiency. The study, conducted in a relatively homogeneous French/English bilingual context in Montreal, supports a model in which early bilingual experience enhances the efficiency and inter-hemispheric cortico-cerebellar connectivity of the language network. However, generalization to other linguistic environments and bilingual populations requires caution.

Using whole-brain network analyses in a large sample, the study demonstrates that bilingual brain networks exhibit greater global efficiency when L2 is acquired early, driven by heightened cortico-cerebellar and inter-hemispheric connectivity. The timing of second language acquisition influences intrinsic functional organization, suggesting a more optimized mechanism for achieving L2 skillfulness during periods of high neuroplasticity in early childhood. These findings advance understanding of how developmental language experience affects brain efficiency and organization and highlight cortico-cerebellar pathways as central to bilingual brain architecture. Future research should explore the impact of learning multiple languages, diverse language pairings, and broader sociocultural contexts on these neural pathways.

Generalizability may be limited by the specific linguistic context (French/English bilinguals in Montreal) and potential modulation by factors such as immigrant and socioeconomic status. The analyses rely on resting-state fMRI, describing undirected functional connectivity, which constrains inference about causal pathways. AoA and L2 proficiency were self-reported, which may introduce measurement variability. The retrospective compilation and focus on two languages may not capture effects present in other bilingual or multilingual populations.