Dynamic neuroplasticity of language networks: The intersection of bilingualism and epilepsy

Significance

We do not fully understand how a neurological insult affects a bilingual person’s two languages. In this study, bilingual individuals with temporal lobe epilepsy showed that their first language was supported by both sides of the brain, regardless of seizure focus. In contrast, their second language was more likely to be processed on the side opposite the seizure focus—especially when seizures began around the time the second language was acquired. These findings suggest that the second language may be more flexible and responsive to neurological disruption than the first. In short, the brain organizes each language differently, depending on the timing of language experience and the timing and location of brain injury.

Abstract

Are bilingual language networks flexible enough to dynamically adapt to neurological insult? We examined language lateralization in 24 bilingual and 46 monolingual adults with temporal lobe epilepsy using functional MRI. In a group of primarily early sequential bilingual patients, the first acquired language (L1) showed more bilateral lateralization than in monolingual patients, with no effect of seizure onset laterality. In contrast, the second-acquired language (L2) was more bilateral in the presence of left hemisphere epilepsy and more left-lateralized in right hemisphere epilepsy. Most notably, in left hemisphere epilepsy, seizure onset closer to L2 acquisition was associated with more right-lateralized L2 representation. These findings suggest a compensatory process in which L2 networks strengthen in the hemisphere opposite the seizure focus, potentially reflecting neural adaptation in early bilingualism. Conversely, L1 appears to have less dynamic reorganization in response to neurological insult. Together, these findings highlight the importance of timing in both language experience and neurological stress in shaping language network organization. They support the view that the bilingual brain is not simply the sum of two monolingual systems, but a dynamic and unique system marked by high interindividual variability, in which divergence between languages may emerge under certain experience- and context-dependent conditions.

Neural organization is both strongly stereotyped and surprisingly flexible. Language lateralization is one classic example of robust hemispheric specialization, as by age ten language processing is predominantly governed by the left hemisphere in an overwhelming proportion of individuals. However, this relationship is not absolute and can be modified by lived experience. One common early-life experience that has been reported to modify typical language organization is bi- or multilingual language acquisition (i.e., proficiency in two or more languages) (1, 2). Bilingualism is observed in over 50% of the world’s population, a proportion which is increasing (3) as globalization, migration, and cultural diversity contribute to an increasingly interconnected world. Another factor modifying stereotypical language organization is neuroplasticity in response to neurological insults such as injury or a developmental pathology. How language networks are affected by the interacting effects of lived experience and neurological insult presents an opportunity to investigate mechanisms of neuroplasticity.

In language development, experience with only a single language (i.e., monolingualism) typically results in a strong left-hemisphere dominance for language. On average, bilinguals are similarly left-hemisphere dominant for all languages (4–6) but patterns of language lateralization are more extensive and more variable than those observed in monolinguals (5, 7, 8). Many studies report that bilingual children and adults also engage right-hemisphere homologues (9–17) and/or nonlinguistic areas (7, 11, 18, 19). Right hemisphere involvement has been observed for the first acquired (i.e., L1) (14, 20, 21), second acquired (i.e., L2) (11, 19, 22–27) or both languages (1, 28–31). This reported heterogeneity is likely due to the many intersecting factors modulating bilingual language organization such as language usage patterns (32), the bilingual ecosystem (33), and unique social-affective influences (2, 34). One strong and tractable factor previously related to language organization is age of acquisition (AoA) of L2 (7, 27, 35, 36), where early acquisition coincides with a sensitive period of heightened neuroplasticity. Meta-analyses of behavioral studies found that early bilinguals who acquire both languages before age 6 tend to show more bilateral language organization across languages, whereas bilinguals who acquire L2 later show more typical left-lateralized patterns similar to monolinguals (1, 37). A second notable factor is language proficiency, with more proficient bilinguals typically showing left-lateralized L2 networks that closely resemble those for L1 (38). Taken together, the strong influence of just two of the many dimensions of bilingualism demonstrate how experience modulates the individual bilingual language network (4, 5, 34).

Another driver of nonstereotypical language lateralization is a neurological insult to the left hemisphere, commonly due to a stroke, head trauma, tumor, or ongoing seizures (39), especially when disruption occurs early in development (40). This reduction in left hemisphere dominance and increased reliance on the right hemisphere is often attributed to adaptation to rely more on healthy tissue to maintain function. However, our understanding of these putatively protective mechanisms remains limited, in part because pathology-related neuroplasticity has been primarily studied in monolinguals. In bilinguals, who show evidence of weaker baseline language lateralization (1), a more distributed network may offer greater capacity for neuroplastic adaptation following neurological insult. It is equally unclear whether such adaptation generalizes across languages (i.e., whether L1 and L2 are equally responsive to insult). The few existing studies in neurological conditions—mostly in poststroke aphasia and neurodegenerative diseases like Alzheimer’s—have reported both shared and distinct effects of pathology on a bilingual’s languages (41–46). For example, L2 has been found to show weaker left-lateralization than L1 following stroke (47), possibly reflecting greater neuroplastic potential or a more distributed cortical representation. Recovery patterns in bilingual aphasia are also heterogeneous: Some individuals show greater impairment in L1, others in L2, and others recover both languages in parallel (42). These discrepancies likely reflect complex interactions among the nature and timing of neurological insult, specific demands on language control networks, and heterogeneity in bilingual profiles, including varying AoA, proficiency, and language pairings (48). As a result, theoretical tension remains around whether and how a bilingual’s languages differentially adapt to pathology.

Studying bilingualism in temporal lobe epilepsy (TLE) offers an opportunity to gain insight into these questions. TLE disrupts temporal lobe regions in language networks and often arises in childhood or young adulthood. In a previous case series of adults with TLE (49), we used functional MRI (fMRI) to compare language lateralization between monolinguals’ L1 and bilinguals’ L2 matched on proficiency. We found that, compared to both monolinguals and healthy bilinguals, bilingual patients showed greater bilateral L2 lateralization in the presence of left (but not right) hemisphere epilepsy. We suggested that bilingualism, in the context of left hemisphere pathology, might enhance neuroplasticity in language networks (49). However, that study was limited by its small sample size, inability to examine L1 lateralization, reliance solely on visual language encoding tasks, and the confounding effect of left-handedness—a factor known to independently affect language lateralization. Additionally, the underlying factors contributing to greater L2 bilaterality in bilinguals with left TLE were not explored.

Current Study.

The present study builds on our prior work by leveraging an independent cohort of adults with epilepsy to ask: How does an early-life experience like bilingualism interact with a neurological insult to shape neuroplasticity in L1 and L2 language networks? We examined fMRI-based language lateralization in highly proficient (and mainly early) bilingual and monolingual adults with TLE. By including participants with left or right hemisphere epilepsy and seizure onset ranging from childhood to adulthood, we were able to examine how both the location and timing of neurological insult interact with early bilingual experience to shape language organization. We hypothesized that: 1) We would replicate the bilingual effect of increased bilaterality in both L1 and L2 language networks, regardless of the hemisphere of seizure focus, based on findings from neurotypical early bilinguals (1); 2) The L2 language network may be more responsive to neurological insult than L1, given prior evidence that early-acquired L2 is more widespread (7) and potentially more plastic; 3) Individuals whose seizure onset occurred closer in time to L2 acquisition would show greater reorganization away from the hemisphere of seizures, reflecting heightened neuroplasticity during language-sensitive development periods; and 4) Lateralization away from the hemisphere of seizures would be associated with preserved language performance, consistent with adaptive neuroplasticity.

Materials and Methods

Participants.

The sample consisted of bilingual and monolingual adults with medication-resistant TLE undergoing presurgical language fMRI at UCLA between 2012 to 2023. Data were acquired retrospectively through review of existing records. All participants were on antiseizure medications, as is typical in presurgical populations. Patients on medications known to affect language (e.g., topiramate, zonisamide) were routinely titrated off prior to neuropsychological testing in accordance with UCLA’s clinical protocols. Before exclusion, 90 patients had available fMRI data. Inclusion criteria included: 1) age 18 or older; 2) diagnosed with unilateral TLE based on scalp or intracranial EEG; 3) able to complete a language interview and sample fMRI tasks; and 4) fMRI data that passed quality control. Patients were excluded for the following reasons: less than age 18 (n = 2); bilateral (n = 7) or unclear (n = 2) seizure focus; failed fMRI quality control (n = 6); completed only one fMRI task (n = 1); or for bilingual patients, fMRI in only one language (n = 2). This resulted in a final sample of n = 70 for further analysis. This study was approved by the UCLA Institutional Review Board (IRB #10-000405), which granted a waiver of written informed consent for retrospective data use.

Language Interview and Characterization of Bilingualism.

At the time of the fMRI visit and/or neuropsychological evaluation (whichever occurred first), participants were asked whether they self-identify as bilingual or reported fluency in, or active use of, a language other than English. Those who met initial criteria completed a modified in-house version of a Language History Questionnaire which included a list of all languages spoken fluently, AoA, self-reported proficiency in reading, writing, comprehension, and speaking, as well as current use patterns for each language. In addition, participants underwent a language interview addressing various aspects of language comprehension and production through a detailed discussion of symptoms. This interview was conducted in English for all included participants, given that English was reported as the dominant language for 21 patients, with the other three demonstrating sufficient English proficiency. These data informed the examiner’s decision regarding whether language mapping would be conducted in one or multiple languages.

fMRI Language Tasks.

As part of clinical language mapping, participants completed three functional tasks using stimuli from different modalities (Fig. 1): 1) object naming (OBJ) with action generation: Patients silently named black-and-white line drawings, and generated an action verb related to the object; 2) reading responsive naming (RRN): Patients silently named objects in response to three-word written description (e.g., “a long yellow fruit”); and 3) auditory responsive naming (ARN): Patients silently named objects in response to three-word oral descriptions presented via headphones. A detailed description of these tasks is published elsewhere (28, 50) and a modified version of a subset of languages is freely available at www.cogneuro.net/hbm2017. This task design is used to clinically map language at UCLA, was previously shown to detect reliable language function (28, 50), and is among the designs recommended by recent consensus recommendations for clinical fMRI language mapping (51).

Fig. 1.

(A) fMRI task design consisting of 12 alternating blocks of rest (fixation cross) and task, completed for three naming tasks and in two languages for bilingual patients; (B) three naming tasks in the visual (i.e., OBJ and RRN) and auditory (i.e., ARN) modalities; (C) a top-voxel activation approach was used to ensure a less biased approach to create individualized thresholds using the top N% of activated voxels for each patient at multiple thresholds; (D) To derive a single measure of a language network, we combined information from all three tasks by characterizing the functional overlap between task modalities, using a conjunction analysis approach; (E) Laterality (i.e., hemispheric asymmetry) indices were calculated based on conjunction maps for each individual patient and used as the main outcome variable in analyses.

Each task was composed of alternating 10 s blocks of task and rest (n = 12 each; rest block presented first). Six patients with lower cognitive ability (e.g., a globally suppressed cognitive profile due to a developmental delay or aphasia as evidenced by medical chart and/or determined by a neuropsychological evaluation), completed the same tasks but at a slower pace. For this version, rest blocks were 10 s whereas task blocks had the following timing: OBJ: 12.5 s; reading: 20 s; auditory naming: 15 s. We conducted sensitivity analyses excluding these participants (see SI Appendix—p. 3).

Bilingual patients completed the tasks in each language, with the reported dominant language tested first. Though patients sometimes completed more than one run per task due to excessive movement or poor compliance, a single run was selected per task based on a combination of data quality (e.g., motion, alignment) and clinician notes.

Image Acquisition.

Imaging data were acquired on a Siemens Magnetom Prisma (n = 62) or Allegra (n = 8) 3T scanner with a 12-channel head coil. The sequence of acquisition included a conventional three-plane localizer and functional echo-planar imaging (EPI) scans (96 volumes; TR = 2,500 ms, TE = 30 ms; FOV = 200 mm; flip angle = 90°; 34 slices; matrix = 64 × 64 × 34, slick thickness = 3.1 mm). High-resolution structural T2-weighted EPI volumes were acquired for each set of fMRI tasks for use during clinical language mapping but were not used for current analyses (see below).

Image Processing.

fMRI data processing was carried out using AFNI (52) and SUMA (53). Preprocessing was carried out using defaults in afni_proc.py with some modifications: slice-time correction, nonlinear warping of BOLD and participant EPI template to a MNI EPI template using a validated approach (54) (i.e., “EPInorm” approach), motion correction using 3dvolreg, smoothing (4.0 mm; full-width at half-maximum Gaussian kernel), and scaling each voxel’s time series to a mean of 100. The participant EPI template was derived by taking the average of the BOLD run. MNI warp and motion correction were applied in a single step. Runs with gross misalignment, gross anatomical warping, or more than 30% of volumes censored were excluded; runs were also excluded if any contraindications were written in scan notes (e.g., patient did not comprehend instructions, did not hear audio, or fell asleep).

fMRI Data Analysis.

First-level analyses were performed with AFNI’s 3dDeconvolve, using the BLOCK response function. Standard motion measures from AFNI’s 3dvolreg (rotation, translation, and their first-order derivatives) were modeled as nuisance regressors. Task and rest were modeled as regressors of interest. A general linear test of task versus rest was modeled to identify voxels that were more active during lexical-semantic processing: The subsequent t-maps were used for all further analyses. To examine the degree of functional overlap between task modalities (i.e., OBJ, ARN, and VRN), we utilized a conjunction analysis approach which has the benefit of attaining robust language-specific activation while eliminating any task-specific (e.g., sensory) activation. This approach has been validated in general (55) and using the same paradigm (28, 50). T-maps were thresholded using a top-voxel activation approach (only positive t-values were considered).

Top-voxel activation approach.

A top-voxel approach was used as a less biased way to create individualized thresholds using the top N% of activated voxels for each patient, rather than setting a threshold for individual t-values. This has the benefit of reducing intersubject variability, increasing the reliability of spatial distribution within subjects across runs (56), and balancing the trade-off between having voxels that respond robustly to lexical-semantic processing while having a sufficient number of voxels in each hemisphere of each patient. This approach has been previously validated and routinely employed in similar studies (56–59). First, a voxel cutoff (N) was derived by taking the median number of activated voxels across all t-maps (i.e., across all participants and all tasks), that survived both a given P-value (P < 0.01, 0.005, 0.001, 0.0005) and a minimal cluster threshold of four voxels. Then, for each t-map, a thresholded map was created by retaining the top N voxels. We used the most stringent threshold (i.e., P < 0.0005) for the main analyses. As sensitivity analyses, we examined effects at each of the other three thresholds, as well as their average (SI Appendix).

To create a conjunction map for each subject across all three modalities, binarized and thresholded t-maps were scaled to a power of 2 (i.e., values of 1: object; 2: auditory; 4: reading) then summed into a single conjunction map. The value at each voxel then represents the combination of modalities that were “active” at that voxel (e.g., 1 for OBJ, 3 for OBJ and ARN, 7 for all three tasks, and so on).

Laterality index calculation.

To examine effects of laterality, for each modality combination in a given conjunction map, the number of voxels in each of the left and right (whole) hemispheres was counted, and a laterality index (LI) was derived using the following formula: LI = [(Left – Right)/(Left + Right)] × 100. Positive LI indicates a leftward asymmetry in activation (i.e., left-lateralized language), whereas negative LI indicates rightward asymmetry. Though we used a continuous LI in all analyses, cut-offs of +0.2 and –0.2 have been routinely used in previous literature to label subjects as primarily left or right-lateralized, respectively, with values in-between considered bilateral. The LI from the conjunction of all three tasks was preferred; if all three were not available, the conjunction of one auditory and one visual task was used.

Neuropsychological Language Measures.

As part of a presurgical evaluation, a subset of patients completed a comprehensive neuropsychological test battery. For this study, we included the following English-language measures: 1) the Vocabulary subtest of the WAIS-IV which assesses the breadth of expressive vocabulary (using age-adjusted scaled scores); 2) a letter fluency task and 3) a category (i.e., semantic) fluency task, completed as part of either the Delis-Kaplan Executive Function System or the Expanded Halstead Reitan Battery.

A subset of bilingual patients also completed the verbal fluency tasks in their other language, allowing for quantitative comparisons of relative proficiency and language dominance (Tables 1 and 2). A bilingual dominance index was calculated based on the total number of correct words produced in L1 relative to the total number of correct words produced in L1 and L2 using a published formula: L1/(L1 + L2) (60), separately for letter and category fluency. A lower index reflects relatively higher L2 proficiency (i.e., 0 = fully L2-dominant), a higher index reflects greater L1 proficiency (i.e., 1 = fully L1-dominant), and a score of 0.5 signifies a perfectly balanced bilingual. For bilinguals completing fluency in both languages, only responses in the target language were scored as correct; code-switches were excluded from total scores. For ease of comparability and given different test versions and categories between participants, demographically corrected standardized scores were used (i.e., T-scores) based on normative data obtained from the respective test manuals.

Table 1.

Demographic and clinical characteristics

	Bilingual		Monolingual
	Bi-LTLE	Bi-RTLE	M-LTLE	M-RTLE	P
n	12	12	27	19
Age (yrs)	31.3 (12.6)	32.1 (11.5)	37.6 (11.4)	32.2 (8.2)	0.22
Education (yrs)*	13.0 (2.3)	13.8 (1.9)	14.0 (2.3)	14.7 (2.8)	0.34
Sex (female)	5 (42%)	4 (50%)	11 (41%)	6 (32%)	0.89
WAIS-IV Vocabulary	10.4 (3.4)	10.0 (3.7)	10.7 (2.2)	10.1 (2.4)	0.94
Handedness (right)	10 (83%)	11 (92%)	23 (85%)	17 (89%)	0.95
Age of seizure onset†	21.5 (12.6)	22.0 (14.8)	18.7 (11.2)	20.1 (10.7)	0.84
Duration of epilepsy†	9.8 (10.4)	10.1 (6.5)	18.8 (15.1)	11.6 (9.1)	0.05
MTS (yes)‡	4 (33%)	5 (42%)	7 (26%)	4 (21%)	0.62
Seizure location (mesial/lateral/both/unclear)	7/3/0/2	8/4/0/0	12/8/2/5	7/10/1/1	0.50
Bilingual variables§	Bi-LTLE	Bi-RTLE	P
L2 AoA	5.7 (5.1)	5.0 (3.3)	0.71
Duration of L2 use	25.4 (9.6)	27.9 (9.5)	0.56
Interval L2 AoA & seizures	14.3 (9.5)	17.7 (13.8)	0.52
Dominance index (category)¶	0.45 (.09)	0.38 (0.07)	0.16
Dominance index (letter)¶	0.48 (0.14)	0.46 (0.06)	0.73
L1 category fluency	37.8 (10.0)	30.9 (10.8)	0.16
L2 category fluency	43.0 (11.4)	48.0 (9.4)	0.36
L1 letter fluency	35.4 (14.9)	35.6 (7.4)	0.97
L2 letter fluency	33.1 (12.2)	42.5 (6.2)	0.07

Note: Bi = bilingual; M = monolingual; LTLE = left temporal lobe epilepsy; RTLE = right temporal lobe epilepsy; L1 and L2 refer to the order of languages learned as reported by bilinguals (i.e., L1 = First acquired language; L2 = Second acquired language); AoA = Age of acquisition; MTS = mesial temporal sclerosis; Vocabulary and letter and category fluency are expressed as standardized scores derived from a demographically matched normative sample (i.e., as scaled scores and T-scores, respectively).

^*Missing for 2 M-LTLE and 2 M-RTLE.

^†Missing for 1 M-RTLE.

^‡Missing for 1 M-LTLE.

^§Based on a subset of bilinguals with available data (Table 2). Verbal fluency tasks were drawn from the Delis-Kaplan Executive Function System (F-A-S/animals and boys’ names) or Expanded Halstead-Reitan Battery (F-A-S/animals). A subset of bilinguals completed fluency tasks in their other language (for most participants this was “P-M-R” and “animales” in Spanish; see Table 2).

^¶A bilingual dominance index was calculated as the number of words produced in L1 relative to the number of total words in both L1 and L2 based on a published ratio: L1/(L1+L2), separately for letter and category fluency. An index of 0 reflects complete L2 dominance, an index of 1 reflects complete L1-dominance, and an index of 0.5 signifies a fully balanced bilingual.

Table 2.

Characteristics of the bilingual sample

ID	Seizure onset	Age	Sex	Edu	Hand	Age seizure onset	L1	L2	L2 AoA	Duration L2 use	English vocab	L1 cat fluency	L2 cat fluency	L1 letter fluency	L2 letter fluency
1	L	38	M	16	L	1	English	Spanish	3	35	17	50	–	60	–
2	L	20	M	13	R	18	Spanish	English	3	17	9	30	53	32	37
3	L	43	M	13	R	38	English	Spanish	–	–	–	–	–	20	–
4	L	38	M	12	L	15	Spanish	English	5	33	6	31	43	39	50
5	L	37	M	12	R	31	Spanish	English	4	33	8	29	37	35	33
6	L	20	F	14	R	14	English	Spanish	4	16	10	50	37	23	20
7	L	29	F	14	R	21	Spanish	English	4	25	10	33	57	20	47
8	L	16	M	11	R	8	Spanish	English	5	11	–	33	57	33	27
9	L	25	F	14	R	21	Spanish	English	6	19	8	27	33	28	30
10	L	28	F	17	R	24	English	Vietnamese	3	25	14	37	–	57	–
11	L	60	F	8	R	47	Spanish	English	20	40	–	39	27	34	30
12	L	21	M	12	R	20	English	Hebrew	–	–	12	57	–	53	–
13	R	27	M	12	R	16	Spanish	English	4	23	5	34	47	39	47
14	R	23	F	13	R	14	Arabic	English	–	–	–	–	–	–	–
15	R	37	M	16	R	29	Spanish	English	3	34	14	20	–	29	–
16	R	22	F	15	R	1	Mandarin	English	4	18	5	39	53	–	40
17	R	42	M	12	R	36	Spanish	English	4	38	–	–	–	–	–
18	R	23	M	12	L	1	Spanish	English	5	18	13	25	33	32	43
19	R	22	F	12	R	17	Arabic	English	1	21	14	–	43	–	50
20	R	39	M	16	R	34	Spanish	English	3	36	10	49	60	45	43
21	R	32	F	12	R	16	Spanish	English	6	26	11	29	47	43	40
22	R	61	M	16	R	53	Korean	English	14	47	–	42	–	42	–
23	R	32	M	14	R	23	Spanish	English	6	26	–	22	60	26	47
24	R	25	M	16	R	24	Spanish	English	5	20	8	20	40	29	30

Note: L1 and L2 refer to the order of the languages learned as reported by bilinguals (i.e., L1 = First acquired language; L2 = Second acquired language); AoA = Age of acquisition; English vocabulary scores expressed as scaled scores (mean = 10; SD = 3); Fluency expressed as T-scores (mean = 50; SD = 10); Cat = category; see Materials and Methods and Table 1 footnote for details.

Statistical Analysis.

Group differences in demographic and clinical variables were tested with ANOVAs and Fisher’s Exact Tests. A mixed ANOVA was used to examine laterality index (outcome) as a function of Seizure Laterality (left versus right; between-subject) and Language (L1 versus L2; within-subject). Additional ANOVAs with Group (bilingual versus monolingual) and Seizure Laterality (left versus right) were carried out for L1 and L2 separately. Simple main effect tests using the pooled variance were performed for significant interactions. Main assumptions of ANOVA were met including normal distribution of laterality indices (Shapiro–Wilk ps > 0.05) and equality of variances (Levene’s ps > 0.05), which is important given unbalanced sample sizes. Kurtosis and skewness were considered within normal limits (<±1) and q-plots were deemed acceptable. Pearson correlations with bootstrapped 95% CI examined relationships between laterality index and clinical/bilingual variables. ANOVAs with laterality index as the dependent variable compared groups with intact versus impaired verbal fluency performance.

Results

Sample Characteristics.

Our final analysis included 24 bilingual patients (12 with left TLE [Bi-LTLE] and 12 with right TLE [Bi-RTLE]) and 46 monolingual patients (27 with left TLE [M-LTLE] and 19 with right TLE [M-RTLE]). Groups did not statistically differ on demographic and clinical variables, including handedness, age, sex, education, age of seizure onset, presence of mesial temporal sclerosis, or English vocabulary. There was a statistical trend (P = 0.050) for M-LTLE having a longer duration of epilepsy compared to the other groups (post hoc ps < 0.05)—Table 1.

Bilingual characteristics (Table 2).

Most bilinguals learned L2 at a young age (M = 5.3 y; SD = 4.2; range 1 to 20), with duration of L2 use averaging 27 y (SD = 9.4; range 11 to 47). Using a cut-off of <7 y old as guided by the literature (7, 61), 90% and 91% of Bi-LTLE and Bi-RTLE, respectively, were early bilinguals. Most bilingual patients learned English as their L2 (79%) followed by Spanish (13%), with English as the most common self-reported dominant language (88%). All patients spoke alphabetic languages, except one Bi-RTLE who spoke Mandarin as their L1. Based on the bilingual dominance index available for a subset of patients, bilinguals were relatively balanced across languages (i.e., index near 0.5). Bi-LTLE and Bi-RTLE did not significantly differ in the bilingual dominance index, L1 and L2 verbal fluency, English vocabulary, L2 AoA, duration of L2 use, or the interval between L2 AoA and age of seizure onset—Table 1.

Whole-Brain Conjunction Group Maps.

We generated group-level conjunction maps across the three naming tasks. As expected, the combined group map revealed robust activation in perisylvian regions, primarily in the left hemisphere, across all patients (Fig. 2). Whole-brain maps stratified by language and seizure laterality group are presented in SI Appendix, Fig. S1. These group-level maps are provided for illustrative purposes only, as all subsequent analyses were conducted at the individual-subject level (i.e., based on laterality indices extracted from native space).

Fig. 2.

Whole-brain group activation map of the conjunction analysis for the full sample (n = 70). Note: For parsimony we included bilinguals’ L1 only. Maps for each of the six groups are provided in SI Appendix, Fig. S1. Two and three-task combinations of conjunctions are depicted with unique colors. Group maps presented in this figure for illustration purposes were thresholded by taking the top 6,657 voxels, which was the median surviving voxel count across all subjects’ task t-maps after applying a voxel-wise threshold of P < 0.01 and a minimum cluster size of 4.

Relationship between Seizure Laterality and Language Lateralization in L1 versus L2.

Fig. 3 plots individual data points and group means of laterality indices. A 2 × 2 model with laterality index as the dependent variable revealed a significant Language × Seizure Laterality interaction of a large effect size [F(1, 22) = 4.9; P = 0.04; η_p² = 0.18] with nonsignificant main effects (ps > 0.05). Follow-up comparisons demonstrated that whereas lateralization did not differ between left versus right seizure onset for L1 (P = 0.90), this difference was significant and of a large effect size for L2 [F(1, 22) = 6.1; P = 0.02; η_p² = 0.22]. That is, bilinguals with left hemisphere seizures showed more bilateral or right-lateralized L2 lateralization (M = 0.11; 95% CI = [−0.12, 0.33]) than those with right hemisphere seizures, who showed typical left lateralization (M = 0.48; 95% CI = [0.26, 0.70]). To ensure this finding was not explained by handedness given its well-known association with language lateralization, we reexamined this interaction controlling for handedness. The interaction remained significant and of a large effect size [F(1, 21) = 4.6; P = 0.04; η_p² = 0.17].

Fig. 3.

(A) Box plots showing individual laterality indices and (B) means (and SE) of laterality indices for each group. Note: More positive laterality index indicates leftward language lateralization. Left and right TLE are depicted by circles and triangles, respectively. Bi = Bilingual; M = Monolingual; LTLE = left TLE; RTLE = right TLE.

Sensitivity analysis across multiple thresholds.

In an examination of the above effects across multiple top-voxel thresholds, the interaction remained significant and of the same effect size when averaging across the four P-value thresholds [F(1, 22) = 4.6; P = 0.04; η_p² = 0.17]. This interaction was significant or marginally significant with similar effect sizes for each of the three other thresholds (SI Appendix, Table S1).

Sensitivity analyses controlling for language dominance and language-specific effects.

Given that language dominance varied across bilinguals, we reexamined the interaction of interest in several ways. First, we repeated the 2 × 2 model, adding a covariate for each patient’s dominant language (i.e., L1 or L2; categorical), which was determined objectively whenever category fluency scores were available for both languages and by self-report for those without fluency. With this covariate, the interaction remained significant and of a large effect size [F(1, 21) = 5.7; P = 0.03; η_p² = 0.21]. A second analysis included only L2-dominant bilinguals (n = 16) and revealed a marginally significant interaction with a similar effect size [F(1, 14) = 4.0; P = 0.07; η_p² = 0.22]. A third analysis controlled for bilingual dominance index based on category fluency in a subset of 15 bilinguals and revealed the same pattern [F(1, 12) = 3.7; P = 0.08; η_p² = 0.24]. Finally, because all Bi-RTLEs had English as their L2, whereas some Bi-LTLEs had other languages as L2, we reexamined the interaction in a subset of 19 bilinguals with English as L2. The interaction remained significant and of comparable effect size ([F(1, 17) = 5.0; P = 0.04; η_p² = 0.23], indicating that the effect was not driven by differences in L2 identity across seizure laterality groups.

Relationship between Seizure Laterality and Language Lateralization in Bilinguals versus Monolinguals.

L1.

Examining monolingual L1 versus bilingual L1, there was a significant main effect of Group of a medium effect size [F(1, 66) = 4.3; P = 0.04; η_p² = 0.06], such that bilinguals showed overall more bilateral language representation (M = 0.23; 95% CI = [0.08, 0.37]) than monolinguals (M = 0.41; 95% CI = [0.31, 0.52]). The effect of Seizure Laterality and the interaction were not significant (ps > 0.05). The effect of Group remained significant controlling for handedness [F(1, 65) = 4.1; P = 0.047; η_p² = 0.06].

L2.

For bilingual L2 versus monolingual L1, there was a significant main effect of Seizure Laterality, such that patients with left hemisphere seizures (M = 0.26; 95% CI = [0.15, 0.37]) showed more bilateral language lateralization than those with right hemisphere seizures (M = 0.44; 95% CI = [0.32, 0.56]); [F(1, 66) = 4.9; P = 0.03; η_p² = 0.07)]. This was qualified by a significant Group × Seizure Laterality interaction of a medium effect size [F(1, 66) = 5.1; P = 0.03; η_p² = 0.07]. Specifically, bilinguals showed more bilateral or right-lateralized L2 than monolinguals in the presence of left (F(1, 66) = 7.7; P = 0.01; η_p² = 0.10], but not right hemisphere seizures (P = 0.58). The interaction remained significant when controlling for handedness [F(1, 65) = 5.0; P = 0.03; η_p² = 0.07].

Sensitivity analysis: Examination across multiple thresholds.

Averaging across the four top-voxel cut-off thresholds: for L1, the main effect of Group remained significant [(F(1, 66) = 5.5; P = 0.02; η_p² = 0.08], again with no interaction (P > 0.05). For L2, the Group × Seizure Laterality interaction approached significance, with a similar effect size [(F(1, 66) = 3.9; P = 0.05; η_p² = 0.06]. Similarly, greater bilateral lateralization was observed in bilinguals than monolinguals but only in left hemisphere epilepsy [(F(1, 66) = 7.4; P = 0.01; η_p² = 0.10]. Similar patterns were observed at each top-voxel cut-off (SI Appendix, Table S2).

Effects of Temporal Proximity between L2 AoA and Age of Seizure Onset.

As only L2 was modulated by seizure laterality, we examined the association between L2 laterality index and the interval between the age when L2 was acquired (i.e., L2 AoA) and the age when seizures developed for patients for whom L2 AoA was available (i.e., interval=age of seizure onset minus L2 AoA) - Fig. 4. This correlation was positive and significant for Bi-LTLE: r(10) = 0.74; P = 0.02; 95% CI: [0.20; 0.93], such that shorter intervals between L2 AoA and seizure onset were associated with more bilateral language representation. In Bi-RTLE, this correlation was in the opposite direction as for Bi-LTLE, but not significant: r(11) = −0.18; P = 0.61; 95% CI: [−0.70; 0.47]; though the magnitude was stronger when removing one visual outlier: r(10) = −0.43; P = 0.22; 95% CI: [−0.83; 0.28].

Fig. 4.

Timing between age of seizure onset and L2 AoA is associated with L2 lateralization for bilinguals with left hemisphere epilepsy. The x-axis represents the interval (in years) between age of L2 acquisition (AoA) and age at seizure onset (i.e., age of seizure onset minus L2 AoA). Bilinguals with left temporal lobe epilepsy (Bi-LTLE) and right temporal lobe epilepsy (Bi-RTLE) are depicted by green circles and gold triangles, respectively. Positive laterality index (y-axis) indicates more leftward language lateralization. Note: The magnitude of the negative correlation for Bi-RTLE was stronger when excluding one visual outlier, depicted inside a black square: (r(10) = −0.43). *P < 0.05.

To determine whether this relationship was specific to L2, we ran a parallel analysis correlating L1 lateralization with age of seizure onset in each group. These associations were not significant across any of the four groups (ps > 0.05; SI Appendix, Fig. S2). In addition, we examined whether age of seizure onset alone (i.e., without accounting for L2 AoA) was associated with L2 lateralization. These correlations were nonsignificant (ps ≥ 0.15; SI Appendix, Fig. S3), though a moderate positive effect was observed in Bi-LTLE: (r(12) = 0.45; 95% CI [−0.17; 0.81]) and a weak negative effect in Bi-RTLE (r(12) = −0.21; 95% CI [−0.70; 0.42]). This suggests that the relative timing between L2 acquisition and seizure onset is a stronger predictor of L2 lateralization than seizure age alone (Fig. 4).

Reanalysis combining current and previous data.

We combined data from Stasenko et al. (49) with our current data to evaluate whether the same association would be observed across independent cohorts in a larger sample (Fig. 5). Controlling for the effect of site, the association between L2 laterality index and the interval between L2 AoA and seizure onset for the combined Bi-LTLE group remained positive and significant: r(16) = 0.53; P = 0.04; 95% CI: [−0.04; 0.87]. The correlation for Bi-RTLE remained in the opposite direction but not significant: r(18) = −0.27; P = 0.29; 95% CI: [−0.71; 0.22], though it approached significance when excluding the visual outlier: r(17) = −0.46; P = 0.08; 95% CI: [−0.77; 0.11].

Fig. 5.

Associations between L2 fMRI laterality index and the interval (in years) between age of L2 acquisition (AoA) and age at seizure onset (i.e., age of seizure onset minus L2 AoA) combining data from the current sample (solid points) and a reanalysis from Stasenko et al. (49). Bilinguals with left temporal lobe epilepsy (Bi-LTLE) and right temporal lobe epilepsy (Bi-RTLE) are depicted by circles and triangles, respectively. The light gray circles depict bilinguals with a late (AoA > 7 y old) L2 acquisition. Note: A visual outlier is denoted by a black square. The magnitude of the negative correlation for Bi-RTLEs was stronger and approached significance when excluding this outlier (partial r(17) = −0.46). *P < 0.05; ^P < 0.10.

Sensitivity analysis: Examination of proficiency/language dominance.

To assess whether proficiency could account for the above pattern observed in Bi-LTLE, we reexamined the proximity correlation, controlling for a proxy for L2 proficiency (i.e., category fluency) in eight Bi-LTLEs. The association between L2 lateralization and the interval between AoA and seizure onset remained of the same magnitude (partial r(8) = 0.72; P = 0.07). Second, we examined bivariate associations between L2 lateralization and 1) L2 letter and category fluency and 2) bilingual dominance indices. None of the metrics were significantly associated with L2 lateralization (rs = −0.02 to −0.14; ps > 0.05; SI Appendix, Fig. S4), suggesting that proficiency does not provide a strong explanation for the observed laterality pattern in Bi-LTLE.

Association between Contralateral L2 Lateralization and L2 Function.

Finally, we tested whether the direction of L2 lateralization—specifically, activation contralateral to the seizure focus—was functionally beneficial, as predicted by neuroplasticity frameworks that view contralateral reorganization as compensatory. We analyzed letter and category fluency performance in L2 (the only language tests available in this cohort) in a subset of 16 bilinguals with available data, collapsing across seizure laterality to increase power (Fig. 6). Participants were classified as impaired or intact based on a T-score >1.3 SD below the normative mean—a commonly used clinical cutoff. To evaluate the relationship between L2 laterality and performance, we recoded the laterality index as (ipsilateral – contralateral)/(ipsilateral + contralateral), such that positive values now indicate more ipsilateral L2 activation (i.e., pathological hemisphere) whereas negative values indicate more contralateral activation.

Fig. 6.

Associations between L2 lateralization and L2 language function (i.e., letter and category fluency) in a subset of bilingual patients, separately by intact versus impaired performance based on a cut-off of <1.3 SD below the normative mean. Ipsilateral refers to the pathological hemisphere (i.e., the same side as the seizure focus); contralateral refers to the nonpathological hemisphere (i.e., opposite side to the seizure focus).

Letter fluency.

A significant effect of group ([F(1, 14) = 6.7; P = 0.02; η_p² = 0.33], revealed that patients with intact fluency showed greater contralateral L2 lateralization (M = −0.39; 95% CI: [−0.71; −0.07]), whereas those with impaired fluency showed more ipsilateral L2 lateralization (M = 0.19; 95% CI: [−0.17; 0.56]).

Category fluency.

Category fluency, however, did not reach statistical significance [F(1, 14) = 3.2; P = 0.096; η_p² = 0.19)].

Discussion

We investigated whether early bilingualism—an experience-dependent factor—interacts with a neurological insult to promote neuroplasticity of language networks. We examined language lateralization patterns in a well-characterized sample of bilingual and monolingual adults with TLE undergoing presurgical language fMRI. With regard to our hypotheses we found: 1) Bilingual L1 was more bilateral than monolingual L1, but hemisphere of seizures did not modify L1 language lateralization; 2) In contrast, L2 was sensitive to neurological insult, with left hemisphere seizure onset associated with more bilateral (i.e., rightward) lateralization whereas right hemisphere seizure onset was associated with more leftward lateralization; 3) Most notably, closer proximity in time of L2 acquisition and age of seizure onset was linked to more bilateral or rightward organization, with an opposite (albeit weaker) effect observed in right hemisphere epilepsy; and 4) Greater lateralization of L2 away from the hemisphere of seizures tended to be associated with better language function, providing preliminary support for the functional relevance of the activation patterns. Together, these findings suggest that early bilinguals’ languages may be differentially neuroplastic in response to neurological insult, with L2 showing greater sensitivity to timing and laterality of pathology. More broadly, this highlights the dynamic interplay between a neurological insult and language development, providing insight into the brain’s capacity for neuroplasticity.

L2 Neuroplasticity and the Proposed Role of AoA.

Our most important finding is the neuroplastic potential of a bilingual’s second learned language in a group of primarily early sequential bilinguals (i.e., AoA before age 7 in >90% of the cohort). At a group level, left-hemisphere pathology was associated with a bilateral or rightward L2 language network and right-hemisphere pathology was associated with a leftward L2 language network. Critically, this had an individual-level component for bilinguals, especially with left-hemisphere pathology, such that the magnitude of the rightward shift in early bilingual left TLE was modulated by the timing of L2 acquisition in relation to the onset of seizures. That is, the near-simultaneous onset of a neurological insult in the language-dominant hemisphere during early childhood L2 acquisition appears to promote a right-lateralized network, insult in the decade following acquisition leads to a bilateral network, and insult in adulthood leads to a more common left-lateralized network (Fig. 7). While the proximity effect between L2 AoA and seizure onset was most evident in Bi-LTLE, a similar but weaker trend was observed in Bi-RTLE in the opposite direction (with a stronger effect when removing a visual outlier). That is, in bilinguals with right hemisphere pathology, a closer proximity tended to be associated with greater leftward L2 lateralization. This RTLE pattern is consistent with our neuroplasticity account, and we hypothesize that the strength of the effect is attenuated due to reduced plasticity potential in the nonlanguage dominant (i.e., right) hemisphere, as well as a restricted range of laterality indices as most RTLEs were left hemisphere dominant. Because the pattern of L2 language organization in early bilinguals appears to be sensitive to the timing of neurological events relative to bilingual acquisition, individual variability in language organization is likely to be high in bilingual patients who suffer neurological insult. We therefore speculate that individual-level interactions like these could explain the heterogenous group-level findings on bilingual language organization as reported previously. This supports the view that bilingual language networks—particularly L2 when acquired early—remain dynamically responsive to insult, and that contralateral recruitment may be a key mechanism of functional preservation during development.

Fig. 7.

A proposed conceptual model of L2 organization for early bilinguals (i.e., AoA < 7 y old) with varying timing of left hemisphere seizure onset. This illustrates the potential interplay between bilingual language experience and pathology-driven neuroplasticity during language-sensitive periods of development. Lightning bolts depict seizure onset timing (childhood, adolescence, adulthood), which interacts with language development to shape L2 lateralization. Empirical findings (column 3) show that earlier seizures are associated with right-lateralized L2, while later seizures are linked to more bilateral or left-lateralized organization in adults. Columns 1 and 2 depict theorized developmental trajectories. The magnitude of language representation is depicted by color and shading (i.e., strongest = bright blue; weakest = white with blue dots).

We observed this dynamic responsiveness in a cohort in which most had an early AoA, leading us to focus our interpretation on early bilingualism. We hypothesize that greater involvement of the contralateral hemisphere is related to a window of plasticity during a language-sensitive (i.e., “critical”) period (62), typically before the age of 7 to 10—a time when language lateralization is still developing and when typically developing children show more bilateral representation (63, 64), as the brain undergoes key maturational changes. Two neurobiological models offer potential underlying mechanisms of this contralateral pattern. The Reorganization model posits that during this highly neuroplastic window, a contralateral compensatory process may be triggered by neurological insult to help maintain language function. In this view, L2—because it is acquired later and may retain greater plasticity than L1—may be more likely to reorganize across hemispheres when seizure onset affects the left hemisphere early in development. A second, developmentally informed interpretation is the Weak Shadow (57) model [also “Weak Echo” (65)], which suggests that the right hemisphere supports language in early development, but that this support typically diminishes as the left hemisphere becomes dominant. If L2 acquisition occurs around the time of left hemisphere seizure onset, the right hemisphere may assume a larger role in language processing—not as a compensatory shift per se, but rather as an extension of already existing early developmental language patterns. At the neural level, these contralateral processes could be mediated by reduced synaptic pruning or decreased transcallosal inhibition, allowing the right contralateral hemisphere to maintain or enhance its role in language during early exposure, when the brain may be more adaptable due to incomplete cortical and corpus callosum development (1). This is supported by evidence of stronger interhemispheric connectivity between homologous regions such as the inferior frontal gyrus in early versus late bilinguals (16). As both models offer plausible explanations, longitudinal studies of pediatric bilingual epilepsy populations are needed to determine whether the observed patterns reflect true compensatory reorganization or developmental scaffolding.

Given a primarily early bilingual cohort, we cannot directly comment on how the observed relationship between the temporal proximity of L2 AoA and age of seizure onset would manifest in those with later L2 acquisition, such as during adolescence or adulthood. Due to our hypothesized mechanisms relying on heightened L2 plasticity, we speculate that for late bilinguals, languages acquired outside the language-sensitive window would not show large contralateral responses to neurological insult. Therefore, L2 may be incorporated into the dominant L1 network (66) and remain left-lateralized even in the context of left hemisphere epilepsy, potentially increasing vulnerability to language impairment without the compensatory benefit of reorganization. However, evidence for this hypothesis is mixed. Studies of late L2 learners shows that neuroplasticity can still occur even after short-term training in adults (2) and that neurocognitive adaptation can remain dynamic and responsive to language use across the lifespan (34). These studies suggest that neuroplastic potential may remain even in later-acquired bilingualism, emphasizing the need for research in this cohort.

Bilingualism and L1 Organization.

In contrast to our L2 findings, we observed that bilinguals overall showed more bilateral L1 lateralization than monolinguals, regardless of the hemisphere of pathology. This suggests that the bilingual brain may support L1 across both hemispheres—perhaps reflecting early-life adaptations to managing multiple languages. Prior meta-analyses of behavioral studies (e.g., dichotic listening) have similarly found that early bilingual experience is associated with more bilateral adult language networks compared to monolinguals, even among sequential bilinguals (1, 37). These findings align with theories of “retroactive” L2-to-L1 influence (1, 37), in which acquiring a second language modifies L1 networks through interactive cross-language feedback (67), and may lead to changes in L1 (68, 69). Our results extend these theories by suggesting that such feedback mechanisms may persist even in the presence of ongoing neurological pathology (28). While L2 showed plasticity in response to seizure laterality, L1 remained consistently bilateral across both left and right TLE, suggesting a more stable and less dynamically reorganizable system. This entrenched organization of L1 may result from early cortical commitment during the sensitive period for first language acquisition, when left hemisphere specialization becomes dominant and increasingly difficult to reverse. Although the right hemisphere may contribute more to L1 in bilinguals than in monolinguals, its privileged and unique status as the first-acquired language may limit its flexibility to reorganize further following later injury. Similarly, monolinguals did not show modulation of L1 lateralization by seizure side. Taken together, while both languages are shaped by bilingual experience, L1 appears to reflect a more developmentally entrenched and less flexible system.

Beyond AoA: Other Experience-Dependent Factors.

Though our explanations emphasize AoA as a key experience-dependent factor shaping language organization, AoA does not operate in isolation. It is embedded within broader models that incorporate sensorimotor integration, language learning context, and social-affective dimensions (35). Other factors such as proficiency, frequency of use, and code-switching also shape bilingual language organization (2, 33, 34, 70, 71). Emerging frameworks conceptualize bilingual outcomes as the product of dynamic interactions between individual cognitive abilities (“expertise”) and the language-learning environment (“ecosystem”) (33). This perspective aligns with our findings, which reflect how neural organization is shaped by the timing of learning and contextual demands. Specifically, our finding that L2 lateralization was most affected when seizure onset occurred near the age of L2 acquisition suggests that nonnative languages may retain greater plasticity and thus may be more susceptible to reorganization when a neurological insult to the left hemisphere occurs during sensitive developmental windows (35).

These experience-dependent accounts diverge from strong interpretations of the spatial convergence hypothesis, which assumes uniform neural representation of L1 and L2 (5, 6). In contrast, the “Adaptive Control Hypothesis” (32) suggests that varying language control demands can lead to functional distinctions within an otherwise convergent system (32, 72). Our findings are more readily explained by experience-dependent models (66, 70), as we propose that spatial convergence is not fixed, but modifiable—shaped by factors such as pathology and timing. Neurological stress, in this context, may shift bilingual language organization from convergence to divergence (71) These findings reinforce the importance of mapping all languages spoken proficiently by bilingual surgical candidates, rather than assuming uniform colateralization as proposed by a strong interpretation of the convergence model.

Though unable to assess the full range of experience-dependent factors, we examined proficiency—one of the most widely studied variables in bilingualism research (38)—in a subset of bilinguals. Using both categorical (L1 versus L2 dominance) and continuous (fluency-based) measures, we found no significant association between proficiency and lateralization patterns. Notably, AoA effects persisted even after accounting for L2 proficiency, lending support for a neuroplasticity-based and developmental timing account of our findings. However, the limited scope of proficiency metrics (i.e., reliance on fluency as a proxy) coupled with a small and relatively homogenous sample (i.e., most bilinguals dominant in L2) limits our ability to draw definitive conclusions about the relative influence of AoA versus proficiency. Future investigations with larger samples will be important to tease apart the influence of these two variables and comprehensively characterize the full spectrum of bilingual language experience (2, 70, 71) to better understand the many possible experience–pathology interactions.

Notably, most bilinguals in the present study were dominant in—and immersed in—their second-acquired language (L2), most often English, which is consistent with a heritage bilingual profile. Therefore, an alternative explanation for a greater effect of left hemisphere pathology on L2 than on L1 is that functional language dominance and language immersion may drive contralateral reorganization in response to seizures, particularly in left TLE. Under this view, the most actively used language (regardless of acquisition order) may be both more vulnerable to disruption and more likely to reorganize when affected. However, this hypothesis does not readily account for the observed time-locked association between laterality and the proximity of L2 AoA to seizure onset, which is most parsimoniously explained by an age-of-acquisition account. Nonetheless, evaluating real-world language use and immersion remains an important future direction for understanding lateralization patterns in bilinguals with epilepsy.

Are L2 Lateralization Patterns Adaptive?

A final question is whether we replicate our previous case-series finding that rightward (i.e., contralateral) L2 reorganization in bilinguals with left TLE was associated with better language outcomes, including naming, letter fluency, and category fluency (49). We interpreted that earlier observation—based on an independent cohort—as suggesting a potentially adaptive shift away from the pathological hemisphere, possibly reflecting beneficial plasticity. In the current cohort, this pattern was only partially replicated. Specifically, we observed a significant relationship only for letter fluency (Fig. 6), such that bilinguals with intact performance showed L2 lateralization away from the seizure focus, whereas those with impaired performance showed L2 lateralization toward it, consistent with our prior finding. However, this association did not extend to category fluency, and a naming measure was not available in this sample. Future studies with larger samples and more comprehensive language assessment (73) will be important to clarify whether the observed L2 lateralization patterns confer a functional advantage in bilinguals with TLE.

Future Directions and Limitations.

Several limitations and future directions merit consideration. Most notably, our sample of bilingual patients was small, particularly for subgroup analyses (e.g., ns = 10 to 11). While replication in an independent cohort strengthens confidence in our key patterns, these correlations should nonetheless be interpreted as preliminary until validated in larger samples. Additionally, the absence of a healthy bilingual control group limited our ability to directly disentangle bilingualism-related effects from epilepsy-related reorganization. Nevertheless, our interpretations were grounded in established findings from the healthy bilingual literature, which served as a theoretical framework for understanding plasticity patterns in the context of neurological insult. The study’s retrospective, observational design also limits causal inference. We hope that these findings will motivate future prospective, longitudinal research to clarify underlying mechanisms across developmental stages. Language function and proficiency was assessed in only a subset of bilinguals and primarily via verbal fluency, which captures a limited aspect of language ability. However, a recent study found that category fluency was highly correlated with a gold standard measure of spoken bilingual proficiency (74). Future studies would benefit from more comprehensive assessments across both languages (e.g., spontaneous speech, naming). Another promising direction involves expanding to bilinguals with greater linguistic distance between languages. Our sample primarily included alphabetic–alphabetic bilinguals, but studies of tonal or logographic languages (e.g., Mandarin) suggest more bilateral or distinct neural networks (75, 76), especially under conditions of greater linguistic distance (76). More detailed neurobiological characterization is also warranted. That is, while we focused on interhemispheric neuroplasticity in core lexical-semantic regions of the inferior frontal and lateral temporal cortex, future work should examine intrahemispheric (i.e., within-hemisphere) shifts and the role of domain-general cognitive control systems (4). Further, our spatial activation analyses may also be complemented by functional connectivity or other network-based approaches (e.g., gradient asymmetries). For example, our group previously demonstrated that bilingualism was associated with changes in white matter connectivity (77) and global network efficiency (78)—factors linked to cognitive resilience in epilepsy.

Finally, these findings have implications for bilingual populations with other neurological conditions including neurodevelopmental (e.g., developmental dysphasia) or acquired (e.g., slow-growing brain tumors, stroke) and neurodegenerative conditions, which may impact language networks at varying stages of bilingual development. A better understanding of language lateralization in these contexts could inform prognosis and rehabilitation strategies for bilingual patients. Our findings suggest that bilinguals with left hemisphere pathology near the time of L2 acquisition may be more likely to preserve L2 function postsurgery due to greater bilateral or rightward representation (i.e., increased brain reserve). This also raises the possibility of developing interventions that target contralateral hemisphere activation (e.g., neuromodulation or speech therapy) to support language recovery in bilingual children.

Conclusion

Our findings underscore the importance of considering how bilingual factors—such as the sequence and age of language acquisition—interact with clinical variables, including the hemisphere of pathology and the timing of neurological insult, to shape language network plasticity. These findings highlight the need for further investigation into how bilingualism influences structural and functional asymmetries in the brain, and how such differences may contribute to cognitive and neural reserve in the context of neurological disease.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Anonymized de-identified data have been deposited in OSF (https://osf.io/uzdek/) (79).