Bilingualism and cognition: The impact of age of acquisition,language use,and proficiency in cognitively unimpaired older adults

Bilingualism, cognitive reserve, and aging

Investigating the potential positive effects of bi- and multilingualism on cognition is crucial for understanding the underlying mechanisms that may protect against cognitive impairment and dementia. This is not only a theoretical question but also has important implications for policies related to healthy aging and the reduction of age-related disorders.^10–12

In recent years, there has been a notable increase in studies examining the cognitive consequences of speaking multiple languages in individuals with neurodegenerative diseases, suggesting that bilingualism may play a protective role by delaying the onset of the disease symptoms. One of the first examples was a study conducted in Canada by Bialystok et al. that showed a 4-year delay in Alzheimer's disease (AD) onset for bilinguals, compared to monolingual peers. ¹³ Craik et al. later confirmed the same time gap in a follow-up study of the same sample. ¹⁴ Other studies followed their work, replicating their results of bilingualism's delaying effect on AD in other cultural backgrounds such as India, ¹⁵ Belgium, ¹⁶ China ¹⁷ and the US. ¹⁸ This effect has been reported in different meta-analyses^19–21 and there is also evidence of delaying effects of bilingualism in mild cognitive impairment (MCI) population.^22–24 Other studies also showed a reduced incidence of MCI ²⁵ and AD ²⁶ in multilingual countries (for a review, see Gallo et al. ⁶ ). However, the bilingual advantage in cognitive decline and dementia prevention has not been consistently observed across all research findings.^27,28

The positive effect of bilingualism against cognitive decline and dementia has often been attributed to the concept of cognitive reserve (CR). According to Stern et al., CR is defined as the discrepancy between the level of age-related neural deterioration and the extent of cognitive impairment. ²⁹ It encompasses the brain's ability to adapt and reorganize through processes such as increased efficiency, capacity, and flexibility of neural networks in response to aging, damage, or disease. These adaptive mechanisms have also been suggested for bilingualism.^5,6,30

However, the relationship between bilingualism and CR remains complex, particularly when considering the interaction between brain health and cognitive performance. Although, some researchers consider that the bilingual advantage may be more evident in older adults, compared to young bilinguals, as age-related cognitive decline provides greater opportunity for bilingualism to generate a protective effect. ³¹ For example, Grant et al. proposed that the previously reported bilingual advantage in episodic memory among older adults, can be explained by enhanced prefrontal function, better preservation of posterior brain regions, and increased connectivity between prefrontal and posterior areas, ³² suggesting that the advantages in episodic memory and executive control among bilinguals may be due to the overlap between brain regions associated with CR and those involved in memory and language.

Meta-analyses show mixed evidence for a bilingual advantage in executive functions among older adults. Degirmenci et al. reported that benefits appear limited to inhibitory control in individuals aged 65–75, with no consistent effects on working memory or task-switching, largely due to variability in language experience, tasks, and participant characteristics. ³³ Similarly, Ware et al. found that the reliability of the bilingual advantage on cognition depends on the type of test employed, suggesting that it is not an all-or-nothing phenomenon. ⁴

While some evidence proposes that bilingualism may contribute to CR, results across the literature remain inconsistent. Therefore, it is necessary to establish the extent to which this advantage exists and how it may be modulated by the specific nature of the bilingual experience, environmental contexts, and other confounding variables.

Bilingual language experience

Different variables associated with bilingualism, including age of acquisition (AoA) of a second language (L2), L2 proficiency, and daily language use, particularly in terms of frequency and switching practices, can influence both language-related processes and executive control,^1,34–36 although their effects vary in magnitude and are not consistently observed across studies.

Regarding AoA, the evidence of its role in modulating executive control is mixed. The distinction between “early” and “late” acquisition is often used as a proxy for the cumulative duration of bilingual experience. For instance, early AoA is linked to better performance in conflict monitoring in children, ³⁷ reduced task-switching costs in young early bilinguals, ³⁸ and in older adults, better performance on working memory task ³⁹ and enhanced episodic recall. ⁴⁰

Language proficiency is another important aspect of the bilingual language experience. It has been shown that higher L2 proficiency predicted better executive control performance^41,42 and is associated with better conflict monitoring. ⁴³ Some studies have found that bilinguals tend to have smaller vocabulary sizes and perform worse in picture naming and fluency tests,^44,45 but in letter fluency tests that require higher executive control demands, high proficiency bilinguals outperformed low proficiency bilinguals. ⁴⁶ The constant need to inhibit a highly proficient competing language may serve as the mechanism that potentially leads to the observed advantages in non-linguistic tasks.^41,47 For example, bilinguals with high proficiency in both their L1 and L2, score higher on executive function tests than those with low proficiency in both languages, while bilinguals proficient in only one language (L1 or L2) score in between these two groups. ⁴⁸

Bilinguals differ not only in proficiency and AoA, but also in language usage, another key variable in defining bilingualism. Contextual factors can lead individuals who acquired two languages early in life to primarily use only one, while late L2 AoA bilinguals may shift their dominant language over time. Therefore, it is essential to examine language usage alongside other factors such as AoA and language proficiency. Dual-language environments where speakers frequently oscillate between languages is associated with enhanced goal maintenance and conflict monitoring. ⁴⁹ Other studies suggest that performance in switching tasks may be influenced by the frequency of daily language switching.^50,51 This distinguishes active bilinguals, who may show enhanced task-switching performance, from passive bilinguals, whose cognitive profiles often resemble those of monolinguals. ⁵²

Although these variables have primarily been explored in younger individuals and the findings remain inconclusive, they are highly relevant to our study, as they will guide our analysis in defining the language profile of participants in our large sample of cognitively unimpaired older adults.

In this study, we investigated the impact of language experience on cognitive aging by analyzing the cognitive performance of cognitively unimpaired individuals aged 45–74 years from the ALFA study, ⁵³ focusing on age of acquisition, proficiency, and language use.

The cohort, recruited in Catalonia (Spain), provides considerable variability in linguistic backgrounds due to the co-official status of Catalan and Spanish. Recent data from the Statistical Institute of Catalonia demonstrate this diversity of language backgrounds: while 34.6% of the population identify Catalan as their L1 and 54.4% identify Spanish, linguistic competence is not uniform across domains. In older adults (55–74 years old) while a high percentage of the population understands Catalan (94.8%), there is a decline in production measures, with only 81.8% able to speak it and 66.8% able to write it. ⁵⁴ Furthermore, these percentages fluctuate significantly based on geography since Spanish dominance is more prevalent in the Barcelona metropolitan area, whereas Catalan dominance is more frequent in interior regions. This sociolinguistic heterogeneity results in a wide spectrum of bilingual profiles, ranging from passive bilinguals with high comprehension but low production, to balanced bilinguals with high proficiency in both languages. This distinction is important, as it allows us to isolate the effects of active language production and switching from mere environmental exposure.

Within this context, three language profiles were defined: a) Early High-Proficiency bilinguals, who acquired both languages in early childhood; b) Late High-Proficiency bilinguals, who acquired Catalan after Spanish but attained high proficiency; and c) Late Low-Proficiency bilinguals, who primarily use Spanish and exhibit limited active use and proficiency in Catalan with a passive exposure to it.

Pure monolingual speakers were not included in this study due to the high levels of cross-linguistic exposure inherent to the region. In Catalonia, Catalan-dominant speakers are necessarily bilingual in Spanish due to its status as the primary language of state administration and media. Spanish was also the mandatory language of schooling during the primary education of most participants in this cohort. On the other hand, Spanish speakers living in Catalonia are consistently exposed to Catalan through social or institutional channels. Therefore, rather than comparing bilinguals to a monolingual control group, this analysis focuses on the meaningful variation within the bilingual experience, operationalized through differences in age of acquisition and proficiency.

The performance of these three language groups was compared across both linguistic and non-linguistic cognitive domains. Specifically, we assessed verbal processing through semantic fluency and episodic memory, and non-linguistic executive functions, including abstract reasoning, visuospatial ability, working memory, short-term memory and processing speed, allowing us to test the hypothesis that the type of bilingualism, conceptualized in terms of age of acquisition (early versus late) and language proficiency (high versus low), may modulate performance on tasks assessing these abilities. Early High-Proficiency bilinguals should outperform Late Low-Proficiency bilinguals, the latter of whom we consider to have a profile more similar to monolinguals. Late High-Proficiency bilinguals will fall in the middle, given the influential role of the L2 AoA.

In particular, we hypothesize that for non-verbal abstract reasoning, visuospatial reasoning, working memory and processing speed, Early High-Proficiency bilinguals would outperform Late Low-Proficiency bilinguals, as higher language proficiency is associated with better executive control, particularly in working memory and conflict monitoring.^41,43 Additionally, AoA also plays a role in cognitive flexibility and conflict monitoring,^37,39,55 and in the active use of the two languages.^41,55 For Late High-Proficiency bilinguals, we expect an effect of later AoA, which might be counteracted by the positive effect of higher language proficiency and active use of the two languages, bringing their performance closer to that of Early High-Proficiency bilinguals.

As for tests with a verbal or linguistic component, we hypothesized that while bilingualism may support the maintenance of episodic memory in older adulthood through its positive effects on executive functioning,^{3,4,6,33,40,41} this advantage would be reduced or absent in tests that place high demands on lexical processing, such as free recall of words, due to bilinguals’ known disadvantages in lexical access and retrieval.^45,46 Additionally, this disadvantage should be more evident for verbal fluency, which relies heavier on a linguistic component. We anticipate that Late Low-Proficiency bilinguals who primarily use one language may perform similarly to, or even better than, highly proficient bilinguals (both early and late) as previous research has shown a bilingual disadvantage in tests involving lexical retrieval^33,39,44,55

Considering the sociolinguistic context of Catalonia, this study seeks to disentangle how specific dimensions of the bilingual experience contribute to better performance in executive control tasks in cognitively unimpaired individuals.

Participants

Participants were selected from the database of the ALFA parent cohort, established by the Barcelonaβeta Brain Research Center. ⁵³ The original aim of the ALFA parent cohort study was to prospectively follow a cohort of cognitively unimpaired individuals, most of whom are offspring of AD patients. To achieve this, the inclusion criteria for enrolment in the study were: (1) being a Spanish and/or Catalan speaker; (2) aged between 45 and 74 years; and (3) providing consent to study procedures and tests, which included a clinical interview, lifestyle and health questionnaires, cognitive assessments, and blood tests for genotyping.

Exclusion criteria were: (1) cognitive performance not meeting the prespecified cutoffs of Mini-Mental State Examination^56,57 (MMSE) < 26, or Memory Impairment Screen^58,59 < 6, or Time-Orientation subtest of the Barcelona Test II ⁶⁰  < 68, or verbal semantic fluency^61,62 (animals) < 12; (2) Clinical Dementia Rating scale ⁶³  > 0; or (3) The presence of a) major psychiatric disorders, b) severe auditory or visual disorders, c) neurodevelopmental and/or psychomotor disorder, d) significant disease that could interfere with cognition, e) neurological disorders or brain injury that could affect participant's cognition and f) suspected pattern of family history of autosomal dominant AD. ⁵³

According to these inclusion and exclusion criteria, 2743 participants were recruited. ⁵³ For the present study, we excluded 81 participants with missing data on the language use questionnaire, 28 participants without L2 AoA information and 129 participants who reported using languages other than Spanish or Catalan. An additional three were removed due to missing age or Semantic Fluency data, resulting in 2502 participants for subsequent analyses.

Data extraction and variable selection

For this study, we extracted the following information from 2502 participants (see Supplemental Table 1 for descriptive statistics of sociodemographic variables, language and cognitive profile of the sample):

Statistical analysis

First, we grouped our sample based on Catalan proficiency and AoA. To define participant groups based on their bilingual language experience, variables with minimal variability were excluded, as they were unlikely to contribute meaningfully to group differentiation. Specifically, Catalan comprehension and reading proficiency, Spanish AoA, and all Spanish proficiency measures.

Group classification was performed using fixed thresholds applied to AoA of Catalan (in years) and Proficiency as the mean self-reported value of Writing and Speaking proficiencies. Participants with mean self-rated Catalan proficiency scores of 3 or lower were categorized as having low proficiency, whereas those with scores of 4 or above were classified as having high proficiency. Based on previous literature,^{33,44,68–70} the age of 6 was chosen as the cutoff for early Catalan AoA, which is when formal compulsory schooling begins in Spain. This criterion therefore includes bilinguals who acquired the language outside a formal learning context. The same cutoff has been used in a previous study with Catalan–Spanish older adults. ²⁴ Those who acquired Catalan at age 6 or later were classified as late bilinguals.

This procedure resulted in four theoretically interpretable language profile groups: (1) Early High-Proficiency Bilinguals (n = 1559); (2) Early Low-Proficiency Bilinguals (n = 87); (3) Late High-Proficiency Bilinguals (n = 537); and (4) Late Low-Proficiency Bilinguals (n = 319). Participants in the Early Low-Proficiency Bilingual group were excluded from further analyses since native exposure is unlikely to result in self-rated low proficiency,^24,33,55 especially in the context of Catalonia due to constant exposure to both Catalan and Spanish.^24,71 Their classification likely reflected data entry or self-reporting errors. See Supplemental Table 3 for demographic and cognitive characteristics of the excluded group. The final sample was 2415 participants (see Supplemental Figure 1).

Second, we conducted an Analysis of Variance (ANOVA) aimed to identify differences in language and sociodemographic variables (age, years of education and adjusted CR) across the three language profile groups. Normality was assessed using the Shapiro-Wilk test, and homogeneity of variances using Levene's test. Due to violations of homogeneity and unequal group sizes, we employed Welch's ANOVA, which is robust to heteroscedasticity and unequal sample sizes, with language profile group as the between-subjects factor. Although traditional ANOVA assumes normality, Welch's ANOVA is robust to moderate violations, particularly in larger samples. Omega squared (ω²) was used to estimate effect sizes. When Welch's ANOVA yielded significant effects, Games-Howell post hoc tests were conducted to perform pairwise comparisons, as this method is specifically designed to handle unequal variances and sample sizes. In addition, a Chi-squared test of independence was conducted to examine potential differences in sex distribution across the three language profile groups.

Third, an Analysis of Covariance (ANCOVA) was conducted to evaluate the effects of language profile group (three levels) and assessment language (Catalan versus Spanish) on test scores. The model controlled for age, sex, years of education and CR. Regarding the language of assessment, the majority of the Early High-Proficiency group (n = 1252; 80.3%) were assessed in Catalan. In contrast, the Late Low-Proficiency group predominantly chose Spanish (n = 265; 83.1%), with only a small subset opting for Catalan (n = 54; 16.9%). Notably, in the Late High-Proficiency group, 41.3% (n = 222) elected to be assessed in Catalan, despite Spanish being their first acquired language. For details, see Supplemental Table 4.

Neuropsychological test scores included in the analysis were (1) Semantic Fluency; (2) MBT Total Free Recall; (3) MBT Total Paired Recall; (4) MBT Total Delayed Paired Recall; (5) MBT Total Delayed Free Recall; (6) WAIS-IV Coding; (7) WAIS-IV Matrix Reasoning; (8) WAIS-IV Visual Puzzles; (9) WAIS-IV Digit Span Total Score; (10) Digit Span Forward; (11) Digit Span Backward; and (12) Digit Span Sequencing. MMSE scores were left-skewed with very low variance, likely influenced by a ceiling effect due to the exclusion criterion (scores < 26). Therefore, MMSE was not included as an outcome variable. To enable direct comparisons across different cognitive measures and groups, z-scores were computed within the whole sample for all test scores.

The ANCOVA model included main effects of language profile group and language of assessment, as well as their interaction (language profile group × language of assessment). Type III sums of squares were used to accommodate unbalanced group sizes and interaction effects. Effect sizes were reported as partial eta squared (η²_p). To control for multiple testing across cognitive outcomes, false discovery rate (FDR) correction was applied to the ANCOVA p-values for each main effect and interaction. Pairwise comparisons of estimated marginal means were then conducted only for effects that remained significant after FDR correction, with Holm adjustment applied to control for multiple comparisons within each set of contrasts using the “emmeans” package in R.

We observed unequal variances (Levene's test, p < 0.05) in MBT Total Delayed Free Recall, WAIS Matrix Reasoning, and Digit Span Sequencing. To address this, we used squared transformation, which most effectively improved variance homogeneity for each affected variable. For MBT Total Paired Recall, none of the transformations successfully corrected variance inequality, and scores were left-skewed. For this variable, we applied robust ANCOVA using heteroscedasticity-consistent standard errors (HC3) via the Anova() function from the car package ⁷² in R, allowing valid inference despite violations of homoscedasticity and normality. For all other variables, Levene's tests were non-significant (p > 0.05), indicating homogeneity of variances.

A post hoc sensitivity power analysis indicated that, given the available sample size (N = 2415), an alpha level of 0.05, and two predictors of interest in the ANCOVA model, the study had 80% power to detect an effect size of f² = 0.004 (corresponding approximately to a partial η² of 0.004). This sample size provided adequate sensitivity to detect even small group differences. All analyses were performed using R software. ⁷³

Sociodemographic variables

For age, we found a statistically significant difference among the groups, F(2, 778.71) = 28.15, p < 0.001, ω² = 0.022, with mean ages ranging from 54 to 57 years and a standard deviation of approximately 6 years. Post hoc comparisons revealed that Late High-Proficiency bilinguals had significantly higher age than Early High-Proficiency bilinguals (p < 0.001) and Late Low-Proficiency bilinguals (p < 0.001) with no significant differences between the later two (p = 0.596). We also found significant differences among the groups in years of education, F(2, 745.09) = 50.14, p < 0.001, ω² = 0.039. Post hoc comparisons showed that both high-proficiency bilingual groups had similar years of education (p = 0.08), but the Late Low-Proficiency bilinguals had significantly lower education than Early High-Proficiency bilinguals (p < 0.001) and Late High-Proficiency bilinguals (p < 0.001). For the adjusted CRQ scores, results showed a significant effect of group, F(2, 742.56) = 44.69, p < 0.001, ω² = 0.035. Post hoc comparisons showed that both high-proficiency bilingual groups had similar adjusted CRQ scores (p = 0.589), but the Late Low-Proficiency bilinguals had significantly lower adjusted CRQ scores than Early High-Proficiency bilinguals (p < 0.001) and Late High-Proficiency bilinguals (p < 0.001). A Chi-squared test of independence indicated that sex was not significantly associated with language profile group, χ²(2, N = 2415) = 2.66, p = 0.264. Overall, significant differences emerged across all sociodemographic variables except sex, with high-proficiency bilinguals showing higher educational attainment and adjusted CR than Late Low-Proficiency bilinguals (Table 1).

OPEN IN VIEWER

Table 1.

Group comparison in sociodemographic and language variables.

Variable	Early high n = 1559	Late high n = 537	Late low n = 319	p-value	ω2	Early high vs. late high	Early high vs. late low	Late high vs. late low
Sex (M/F)	586/973	189/348	130/189	0.264*	-	-	-	-
Age (years)	56.29 ± 6.78	53.9 ± 6.28	55.92 ± 6.08	<0.001	0.022	<0.001	0.596	<0.001
Years of Education	13.64 ± 3.43	14 ± 3.31	11.69 ± 3.5	<0.001	0.039	0.08	<0.001	<0.001
CRQ Adjusted Score	10.47 ± 2.72	10.34 ± 2.52	8.82 ± 2.89	<0.001	0.035	0.589	<0.001	<0.001
Catalan
AoA	1.1 ± 0.56	18.62 ± 8.35	24.44 ± 12.19	<0.001	0.593	<0.001	<.001	<0.001
Comprehension (1–5)	4.97 ± 0.16	4.95 ± 0.23	4.56 ± 0.72	<0.001	0.042	0.038	<0.001	<0.001
Reading (1–5)	4.96 ± 0.21	4.89 ± 0.45	4.24 ± 0.98	<0.001	0.068	<0.001	<0.001	<0.001
Speaking (1–5)	4.95 ± 0.22	4.77 ± 0.45	3.16 ± 0.96	<0.001	0.326	<0.001	<0.001	<0.001
Writing (1–5)	4.78 ± 0.5	4.66 ± 0.56	2.14 ± 0.91	<0.001	0.514	<0.001	<0.001	<0.001
Spanish
AoA	3.22 ± 2.84	1 ± 0.06	1 ± 0.06	<0.001	0.281	<0.001	<0.001	0.916
Comprehension (1–5)	4.96 ± 0.19	4.99 ± 0.11	4.94 ± 0.26	<0.001	0.010	<0.001	0.166	<0.001
Reading (1–5)	4.96 ± 0.19	4.99 ± 0.11	4.94 ± 0.23	<0.001	0.010	<0.001	0.297	0.001
Speaking (1–5)	4.92 ± 0.3	4.99 ± 0.14	4.91 ± 0.34	<0.001	0.022	<0.001	0.916	<0.001
Writing (1–5)	4.93 ± 0.27	4.98 ± 0.15	4.9 ± 0.33	<0.001	0.014	<0.001	0.147	<0.001

Early High: Early AoA High-Proficiency Bilinguals; Late High: Late AoA High-Proficiency Bilinguals; Late Low: Late AoA Low-Proficiency Bilinguals; M: Male; F: Female; AoA: Age of Acquisition; CRQ: Cognitive Reserve Questionnaire. Values are presented as mean (± SD). Post hoc results are for Games-Howell tests. Effect sizes were estimated with omega squared (ω²). An asterisk indicates p-values are for Chi-squared Test (χ ² ). Dash (-) indicates not applicable.

Language profile

Language use

Most of the Early High-Proficiency bilinguals reported a high preference for Catalan during childhood (87%) and adulthood (72%) with some reporting a bilingual preference in adulthood (22%). In contrast, Late Low-Proficiency bilinguals reported a high preference for Spanish during childhood (97%) and adulthood (75%) with some reporting a bilingual preference in adulthood (18%). The Late High-Proficiency bilinguals also reported a high preference for Spanish in childhood (93%) but a more balanced preference between the two languages in adulthood, with 27% having a preference for Spanish, 30% for Catalan and 43% for both languages. These percentages are mean values across all contexts from childhood and adulthood, for specific results of each group in different contexts across their lifespan, see Figure 1 and Supplemental Table 5.

OPEN IN VIEWER

Figure 1.

Percentages of language use responses by language profile group. Late Low: Late Low-Proficiency bilinguals; Late High: Late High-Proficiency bilinguals; Early High: Early High-Proficiency bilinguals.

Cognitive measures

Figure 2 summarizes the significant results of the ANCOVA assessing the effects of language profile group and language of assessment on cognitive measures, while controlling for age, years of education, sex, and adjusted CR scores.

OPEN IN VIEWER

Figure 2.

Adjusted means of cognitive measures with significant main effects of bilingual profile groups. Digit Span Forward significant differences are only in Catalan assessments. AoA: Age of acquisition; MBT: Memory Binding Test; TFR: Total Free Recall; TPR: Total Paired Recall; TDFR: Total Delayed Free Recall; TDPR: Total Delayed Paired Recall. An asterisk indicates a significant difference between groups (FDR-p < 0.05).

For the WAIS-IV Coding test, the main effect of language profile group was significant (F(2, 2405) = 20.23, FDR-p < 0.001, η²_p = 0.017). Post hoc analyses revealed that Late Low-Proficiency bilinguals performed significantly worse than both Early High-Proficiency (t = -4.37, p < 0.001) and Late High-Proficiency bilinguals (t = -3.05, p = 0.005). No significant differences were found between the two high-proficiency groups (t = 1.67, p = 0.094).

Similarly, for Semantic Fluency, a significant main effect of bilingual profile group was found (F(2, 2405) = 7.39, FDR-p = 0.004, η²_p = 0.006). Post hoc comparisons showed that Late Low-Proficiency bilinguals performed significantly worse than Early High-Proficiency (t = -3.74, p < 0.001) and Late High-Proficiency bilinguals (t = -3.54, p < 0.001). Again, no difference was observed between the Early and Late High-Proficiency groups (t = -0.05, p = 0.960).

Additionally, a significant interaction between language profile group and language of assessment was observed for the Digit Span Forward test (F(2, 2405) = 8.74, FDR-p = 0.002, η²_p = 0.007). Post hoc comparisons indicated that when the test was administered in Spanish, there were no statistical differences between the language profile groups (all p-values > 0.05). However, during Catalan assessments, Early High-Proficiency bilinguals scored significantly higher than both Late High-Proficiency (t = 4.36, p = 0.002) and Late Low-Proficiency bilinguals (t = 3.57, p = 0.004). This test also yielded a significant main effect of language of assessment (F(1, 2405) = 11.15, FDR-p = 0.011, η²_p = 0.005), driven by higher scores in Spanish assessments. For more information about demographics and cognitive test scores of the sample grouped by language of assessment (Spanish or Catalan), see Supplemental Table 6 and Supplemental Figures 2–4.

For the remaining cognitive measures of episodic memory (MBT), visuospatial and abstract reasoning (WAIS-IV Visual Puzzles and Matrix Reasoning), and working memory (WAIS-IV Digit Span Total score, Digit Span Backward, and Digit Span Sequencing), there were no significant main effects of language profile group, language of assessment, or their interactions after FDR correction (FDR-p > 0.05). For more details, see Table 2.

OPEN IN VIEWER

Table 2.

ANCOVA results for cognitive measures by language profile group and language of assessment.

Cognitive measure	Main effect	F (df)	p	FDR, p	η2p
Semantic Fluency	Group	7.39 (2, 2405)	<0.001	0.004	0.006
	Language	4.81 (1, 2405)	0.028	0.119	0.002
	Interaction	1.99 (2, 2405)	0.137	0.298	0.002
MBT Total Free Recall	Group	3.31 (2, 2405)	0.037	0.158	0.003
	Language	0.20 (1, 2405)	0.657	0.660	<0.001
	Interaction	0.05 (2, 2405)	0.955	0.955	<0.001
MBT Total Paired Recall	Group	2.07 (2, 2405)	0.126	0.225	0.002
	Language	2.27 (1, 2405)	0.132	0.287	<0.001
	Interaction	2.09 (2, 2405)	0.124	0.298	0.002
MBT Delayed Paired Recall	Group	1.59 (2, 2405)	0.203	0.225	0.001
	Language	1.20 (1, 2405)	0.273	0.287	<0.001
	Interaction	0.81 (2, 2405)	0.443	0.298	<0.001
MBT Delayed Free Recall	Group	1.83 (2, 2405)	0.161	0.233	0.002
	Language	1.47 (1, 2405)	0.225	0.419	<0.001
	Interaction	1.18 (2, 2405)	0.309	0.446	<0.001
WAIS-IV Coding	Group	20.23 (2, 2405)	<0.001	<0.001	0.017
	Language	4.21 (1, 2405)	0.040	0.119	0.002
	Interaction	2.26 (2, 2405)	0.104	0.298	0.002
WAIS-IV Matrix Reasoning	Group	1.64 (2, 2405)	0.193	0.234	0.001
	Language	5.48 (1, 2405)	0.019	0.119	0.002
	Interaction	1.72 (2, 2405)	0.179	0.333	0.001
WAIS-IV Visual Puzzles	Group	0.07 (2, 2405)	0.928	0.928	<0.001
	Language	0.19 (1, 2405)	0.660	0.660	<0.001
	Interaction	0.42 (2, 2405)	0.766	0.847	<0.001
WAIS-IV Digit Span Total	Group	2.37 (2, 2405)	0.093	0.225	0.002
	Language	3.99 (1, 2405)	0.046	0.119	0.002
	Interaction	3.55 (2, 2405)	0.028	0.187	0.003
Digit Span Forward	Group	2.01 (2, 2405)	0.135	0.225	0.002
	Language	11.15 (1, 2405)	<0.001	0.011	0.005
	Interaction	8.74 (2, 2405)	<0.001	0.002	0.007
Digit Span Backward	Group	1.53 (2, 2405)	0.216	0.234	0.001
	Language	0.45 (1, 2405)	0.501	0.651	<0.001
	Interaction	1.99 (2, 2405)	0.136	0.298	0.002
Digit Span Sequencing	Group	2.06 (2, 2405)	0.127	0.225	0.002
	Language	0.94 (1, 2405)	0.331	0.479	<0.001
	Interaction	0.25 (2, 2405)	0.782	0.847	<0.001

Models were adjusted for age, sex, years of education, and Cognitive Reserve. Group: Language Profile Group (Early High, Late High, Late Low); Language: Language of Assessment (Catalan versus Spanish); Interaction: Group × Language; η²_p: Partial Eta Squared; FDR: False Discovery Rate; Early High: Early AoA High-Proficiency Bilinguals; Late High: Late AoA High-Proficiency Bilinguals; Late Low: Late AoA Low-Proficiency Bilinguals. Bold text indicates statistical significance after False Discovery Rate (FDR) correction.

To address the robustness of our group classifications, we conducted two distinct sensitivity analyses modifying the AoA cutoff. In addition to our original cutoff (AoA < 6), we tested a stricter cutoff to capture early preschool home exposure (AoA < 5) and a higher cutoff (AoA < 10). Additionally, we include the same analysis with the Early Low-Proficiency (n = 87) group that we excluded prior to analysis in the main manuscript. We applied the same FDR corrections for multiple comparisons across all models. The significant cognitive advantages observed in Semantic Fluency, WAIS-IV Coding, and the interaction for Digit Span Forward survived FDR correction in all three additional models. Furthermore, tests that showed null effects in our primary analysis remained statistically nonsignificant across all variations (for details, see Supplemental Tables 7–9).

Non-linguistic domain

In the non-linguistic domain, we hypothesized that Early High-Proficiency bilinguals would outperform Late Low-Proficiency bilinguals due to the combined benefits of early AoA, high language proficiency, and active use of both languages. We further hypothesized that Late High-Proficiency bilinguals would show intermediate performance. In this group, the potential disadvantages associated with a later L2 acquisition might be partially offset by the benefits of higher proficiency and frequent language use in adulthood.

Our results suggest that both high-proficiency bilingual groups, regardless of AoA, outperformed Late Low-Proficiency bilinguals on the Coding subtest (WAIS-IV). This task integrates multiple cognitive processes including processing speed, visual-motor coordination, short-term memory and incidental learning. As such, it is difficult to determine whether the combined effects of AoA and language proficiency target a specific underlying cognitive process involved in test performance.

However, performance on tests that directly assess short-term memory suggest a distinct effect of AoA. On that matter, robust group differences were found for the Digit Span Forward subtest, where Early High-Proficiency bilinguals outperformed both Late High-Proficiency and Late Low-Proficiency bilinguals when assessed in Catalan.

The patterns observed in both tests may reflect differences in the specific executive functions each test engages, and in the underlying cognitive processes supporting those functions, as well as the degree of bilingualism needed to attain a measurable cognitive benefit. While the Digit Span test primarily assesses verbal short-term memory, the group differences observed in the Coding subtest may more directly reflect better processing speed rather than broader working memory advantages. Alternatively, rather than attributing the differences solely to specific cognitive processes or benefits, it is possible that the results are better explained by the degree of bilingual experience required to detect an effect. Notably, the Early High-Proficiency bilinguals, who had the highest level of bilingual exposure and history of dual-language use, may be more sensitive to subtle cognitive advantages, thus making group differences more detectable in this subgroup.

Interestingly, no significant differences were observed between the groups on the Digit Span Backward or Sequencing tests. Although high-proficiency bilinguals showed better performance in tests involving speed and attentional control, as evidenced in both the Coding and Digit Span Forward tests, this benefit did not extend to more demanding working memory tests such as Digit Span Backward and Sequencing. This suggests that the cognitive advantage may lie more in attentional processing and speed than in the manipulation and maintenance aspects of working memory.

No significant differences were found in other non-linguistic tests, such as the Visual Puzzles and Matrix Reasoning subtests of the WAIS-IV, which assess non-verbal abstract reasoning and visuospatial reasoning. This suggests that any modulation associated with bilingualism may be selective and specific to certain cognitive domains and tests, rather than reflecting a generalized enhancement of cognition. Indeed, the evidence for a bilingual advantage in these cognitive domains remains mixed. For instance, a meta-analysis reported a bilingual advantage in abstract and symbolic reasoning, however, the studies included primarily young participants. ⁷⁴ Regarding working memory, there are studies reporting small to moderate bilingual advantages in this domain,^74–76 however, other meta-analyses reported a null effect of bilingualism on working memory. For example, Degirmenci et al. found no clear evidence of a bilingual advantage in the updating domain, which is closely related to working memory, in cognitively unimpaired older adults. ³³ Similarly, Lehtonen et al. reported that the already small effect sizes observed in working memory tests disappeared after correcting for publication bias. ⁴⁴ Notably, in one study, monolinguals even outperformed bilinguals on the Digit Span Backward test. ³⁹

Taken together, these results suggest that lifelong use of a second language, including continued use later in life, may modulate cognitive performance, though not uniformly across all domains. The high statistical power of our cohort allows us to interpret these null findings not as a lack of sensitivity but as robust evidence supporting a domain-specific cognitive enhancement. These null results help complete the picture of how the cognitive benefits of bilingualism may operate across the lifespan, as well as clarify the nature of these effects in relation to language control and speech production in bilingualism. For instance, the absence of positive effects on non-verbal abstract reasoning and visuospatial reasoning may be explained by the fact that these functions are more strongly associated with baseline fluid intelligence and general cognitive capacity. This suggests that they are less directly involved in language control processes and in the interaction between linguistic and non-linguistic domains. Additionally, both tasks lack the speeded conflict-monitoring or task-switching demands that are specifically hypothesized to be enhanced by dual-language management. ⁴⁷

Furthermore, this domain-specific pattern aligns with the natural trajectories of cognitive aging. According to Grundy et al. ⁷⁵ as bilingual experience and proficiency increase, the brain shifts from effortful, top-down frontal recruitment to more automatic, subcortical, and posterior processing. This shift results in neural efficiency, where bilinguals achieve the same behavioral scores as monolinguals but do so by recruiting fewer frontal resources. Therefore, null findings in memory and reasoning tasks may simply reflect that bilinguals have optimized their processing to match standard performance levels with a more “economic” neural signature. Due to processing speed being one of the earliest affected cognitive domains in healthy older adults, the protective effects of bilingualism may be most visible here. In contrast, cognitive functions that remain relatively stable during this period may not yet exhibit a measurable difference in our cohort's age range or these specific neuropsychological tests. ⁴ Similarly, verbal episodic memory is more reliant on specific medial temporal structures (e.g., the hippocampus) ⁴⁰ where bilingualism-induced protective effects may only manifest when pathological decline begins in these structures. While some aging processes may have already begun, these participants are still cognitively healthy with no signs of memory impairment.^40,48

Our findings are also consistent with those reported by Gallo et al. regarding the relationship between L2 proficiency and the efficiency of the attentional network in older adults. ⁴¹ The faster performance observed in our high-proficiency groups (Coding and Digit Span Forward) supports the hypothesis that bilingualism may mitigate age-related cognitive decline by enhancing general processing efficiency rather than inhibitory control. ⁷⁷ In older populations, this advantage is often characterized by faster reaction times across both congruent and incongruent conditions ⁶ suggesting a broader framework of attentional control or increased processing speed^3,78 rather than specific executive functioning advantages.

Furthermore, it is important to acknowledge that the observed effect sizes were small and that advantages were confined to a limited number of measures. Therefore, this pattern of domain-specific impact may indicate that the influence of bilingualism on the cognitive profile of cognitively unimpaired older adults, such as those in the ALFA cohort, is relatively subtle, emerging only in specific tasks with specific executive demands.

Linguistic domain and language of assessment

For verbal episodic memory performance, we found no significant differences across language profile groups or language of assessment for any of the MBT measures. These results are in agreement with our hypothesis that potential cognitive benefits do not extend uniformly to all cognitive domains, particularly those evaluated with a test that relies on a verbal or linguistic component. For instance, Fernandez et al. found that bilingual older adults recalled fewer words than age-matched monolinguals in a free recall test, likely due to the high lexical demands of the test. ⁷⁹ To some extent, this supports the idea that a verbal disadvantage can counteract the executive control benefits associated with bilingualism, ⁸⁰ and that when the verbal component is reduced, the bilingual advantage becomes more evident. ⁴⁰

Nevertheless, verbal fluency results were not in line with our hypothesis. Both high-proficiency bilingual groups performed significantly better than the Late Low-Proficiency bilinguals. This finding contrasts with the broader literature, which typically associates bilingualism with disadvantages in lexical retrieval tests. Such disadvantages are often attributed to reduced exposure to each individual language compared to a monolingual speaker, and increased lexical interference from the constantly activated non-target language.^44,81

However, some studies have reported exceptions and mixed findings. ⁸² Costumero et al. found that while both monolingual and bilingual groups experienced cognitive decline, including in semantic fluency, over a 7-month follow-up period, monolinguals showed a greater overall cognitive decline than bilinguals, suggesting more preserved performance in bilingual individuals. ⁸³ In cross-sectional studies, when bilinguals are matched with monolinguals on vocabulary size as a proxy for language proficiency, they either outperform ⁸⁴ or perform comparably to monolinguals.^46,85 Similarly, reported no significant differences in a semantic fluency test between Bengali-English bilinguals and English monolinguals matched on receptive vocabulary, age, education, and non-verbal intelligence. ⁸⁶ In the same way that cognitive benefits associated with bilingualism may be domain- and test-specific, potential disadvantages may also be limited to particular contexts. For example, Bialystok et al. reported that monolinguals produced more words only when compared with low-proficiency bilinguals, while no significant differences were observed between monolinguals and high-proficiency bilinguals. ⁴⁶ Our results exemplify this by demonstrating performance differences within the bilingual population based on their proficiency levels, with Early and Late High-Proficiency groups scoring similarly in the semantic fluency test and outperforming the Late Low-Proficiency bilingual group.

The influence of language of assessment was restricted to tasks with high verbal processing demands. No significant effects were observed in non-linguistic domains, non-abstract reasoning, processing speed, visuospatial reasoning, or working memory. This suggests that while L2 processing may impose a subtle cognitive load during lexical retrieval and verbal short-term memory, it does not interfere with executive function domains.

In the semantic fluency test, participants assessed in Spanish produced more words on average than those assessed in Catalan. However, this effect did not survive FDR correction, and should therefore be interpreted with caution. This trend suggests that lexical retrieval may be facilitated when the test is administered in the participant's dominant language, particularly among late bilinguals. A similar pattern emerged in verbal short-term memory evaluated with the Digit Span Forward test. Prior research has shown better digit span performance in a participants first language ⁸⁷ or language of schooling,^88,89 which for both late AoA bilingual groups was Spanish.

In addition, the interaction between bilingual group and language of assessment was significant in Digit Span Forward. While no significant group differences were found in Spanish assessments, Catalan assessments revealed that Early High-Proficiency bilinguals outperformed both late AoA groups. This suggests that even in relatively simple cognitive tests, verbal processing demands in a bilingual's L2 may subtly hinder performance, particularly when rapid lexical access is required, which can directly impact performance of late AoA bilinguals despite high L2 proficiency, compared to early bilinguals, who likely possess more integrated and automated phonological representations in both languages.

Influence of bilingual language experience on cognitive performance

The results from the verbal short-term memory tests suggest a specific advantage for early bilinguals. Early AoA is typically associated with greater advantages due to increase experience of controlling two languages. For example, in one study, early AoA bilinguals exhibited significantly smaller interference cost on the Flanker test compared to late bilinguals and monolinguals. ⁵⁵ Additionally, the observed benefits of early AoA may result from the combined effects of earlier exposure, higher proficiency and subsequently greater frequency of use across both languages. In a study with Spanish-Catalan bilinguals, higher frequency of language use was associated with better working memory and phonemic fluency, measured as a composite score including Letter-Number Sequencing, Corsi, phonemic fluency, and Digit Span tests. ⁷¹ Interestingly, greater language use was also linked to increased white matter deterioration, yet participants maintained equivalent cognitive performance across domains, suggesting a CR effect. Although they did not include direct measures of language proficiency or AoA, the sample was composed of highly educated Spanish-Catalan bilinguals that likely overlaps with the high-proficiency bilinguals identified in our study.

Our findings suggest that although early AoA is associated with higher language proficiency and active use of both languages, leading to better cognitive performance in domains such as verbal short-term memory, high proficiency itself was also linked to cognitive benefits, even among bilinguals with a late AoA. This implies that language proficiency may have an independent effect on cognition. In our study, participants in the Late High-Proficiency group exhibit a distinct sociolinguistic trajectory: they shift from a strong preference for Spanish in childhood to a more balanced or even Catalan-dominant use in adulthood (Figure 1). In contrast, the Late Low-Proficiency group does not demonstrate this shift, remaining Spanish-dominant despite living in the same dual-language environment. This confirms that high proficiency in both languages is closely linked to a more active bilingual engagement in adulthood, which is expected, as greater language use is likely to support and reinforce language proficiency and vice versa.^{1,9,34,49,90,91} While the relationship may be bidirectional, meaning that high proficiency may promote greater language use, and frequent use may reinforce proficiency, both scenarios suggest a positive feedback loop in bilingual language engagement. A novelty of our study lies in identifying and isolating this specific subgroup of late learners who successfully transition to an active bilingual mode. This provides a unique opportunity not only to examine a profile that authentically reflects the regional sociolinguistic reality, but also to explicitly test whether the increased active use of a second language later in life can still confer protective cognitive benefits. Finally, the specific type of proficiency may also be an important factor. In a previous study, Calabria et al. found that self-reported Catalan proficiency in speaking was the most influential variable when calculating a bilingualism composite score that ranged from passive to active bilingualism. ²⁴ Similarly, production measures of self-rated proficiency in Catalan, rather than comprehension measures, were most strongly associated with cognitive benefits in our study.

In summary, our results indicate that high-proficiency Spanish-Catalan bilingual older adults who actively use both languages showed better cognitive performance in specific domains compared to passive, Spanish-dominant bilinguals. These findings underscore the idea that the degree and nature of bilingual engagement, rather than just being bilingual, are key factors driving these cognitive benefits.^6,24,92–94 Our research contributes to the growing body of evidence that supports the unique and beneficial role of specific aspects of bilingual experience to both cognitive and neural reserve, which have been consistently associated with protective effects against age-related cognitive decline and enhanced resilience in the face of MCI or AD.^5,6,93,95,96 Additionally, our findings suggest that the benefits of lifelong L2 use are domain- and test-specific, highlighting that the bilingual advantage largely depends on specific aspects of language experience and how bilingualism is defined and measured.

Our study has four main limitations. First, self-reported measures of language use and proficiency are subject to bias and may not fully capture bilingual experience. Although some studies have demonstrated that self-rated proficiency is highly correlated with objective measures of proficiency,^9,97,98 future research should combine standardized self-report instruments to capture detailed language history such as The Language and Social Background Questionnaire (LSBQ) ⁹⁹ or the Language Experience and Proficiency Questionnaire (LEAP-Q),^100,101 with objective behavioral measures to mitigate the inherent subjectivity of self-assessments.

Second, bilingualism does not exist in isolation, and its effects on cognition can be confounded or influenced by socioeconomic status^77,102 or other lifestyle factors such as occupational complexity or engaging in physical, intellectual, and social activities as well as music training.^6,70,92,103 In our study, we addressed these potential confounding variables with a CR proxy, ⁶⁷ which includes questions about the educational background, occupational complexity, social and physical activities, and cultural and intellectual pursuits. However, future research should include more detailed measures of socioeconomic status and lifestyle factors to further isolate the effects of bilingualism.

Third, our findings are likely generalizable to bilingual environments that share similar sociolinguistic characteristics with our cohort, but not to all contexts. We highlight several aspects that we consider particularly important. A primary aspect is the presence of bilingual communities in which both languages are widely used in everyday life. This includes contexts where two or more languages are co-official and actively used across social interactions and societal domains, such as mass media (e.g., television, radio) and institutional communication. Furthermore, in our study, we did not formally control for migration status. Given the sociolinguistic demographics of Catalonia, we expect Early High-Proficiency bilinguals to be local native speakers from Catalonia, whereas participants in late acquisition groups are likely composed of individuals who have relocated to Catalonia from other regions of Spain, rather than international immigrants. For this reason, our findings may not fully generalize to bilingual contexts where second-language acquisition is primarily driven by international immigration. Additionally, Spanish and Catalan are typologically similar and belong to the same language family. Although the role of typological distance in modulating cognitive benefits remains unclear, our findings are most directly applicable to contexts involving closely related languages. Future research should aim to directly compare cognitive outcomes across diverse bilingual populations to determine how these environmental and linguistic variables further modulate the bilingual advantage.

Finally, our definition of late AoA encompasses individuals who learned their second language during late childhood or adolescence as well as those who learned it in adulthood. Because language acquisition is generally easier and relies on different neurocognitive mechanisms during childhood compared to adulthood, grouping these diverse learning stages together is a limitation of the current study. Future research utilizing larger samples should further subdivide late bilinguals or treat AoA as a continuous variable to investigate whether learning an L2 in adolescence confers different cognitive benefits compared to learning an L2 later in life.

In conclusion, our findings highlight that the cognitive effects of bilingualism in aging are not uniform but domain-specific, with benefits most evident in tasks involving attentional control and processing speed, rather than across all cognitive domains. Moreover, the extent of cognitive benefits depends on specific aspects of the bilingual experience, such as language proficiency and AoA. These results have both theoretical and clinical implications. Theoretically, they deepen our understanding of the relationship between cognition, bilingualism, and aging within the framework of CR. Clinically, they inform best practices for assessing language and cognitive deficits in bilingual individuals and suggest that second language use may serve as a valuable strategy for preventing cognitive decline and could be potentially integrated into cognitive training programs.