Abstract

Inhibitory control is a core cognitive function that is primarily associated with activation in the prefrontal cortex (PFC) and is the cognitive function that inhibits impulses, thoughts, and suppresses irrelevant information to an identified goal or task. Prior research suggests that bilingualism may affect brain activity related to inhibitory control, yet few studies have compared functional activity between monolingual and bilingual children. The current study used functional near-infrared spectroscopy (fNIRS) to examine region of interest comparisons and task-state functional connectivity across the PFC during an interference suppression Simon task with 13 bilingual (East Asian or Ibero-romance paired with English) and 13 age-matched English monolingual preschoolers. Results showed no significant differences in behavioral measures of interference suppression. However, bilingual preschoolers showed lower oxygenated hemoglobin activation and more localized patterns of connectivity within the PFC, suggesting more efficient processing during suppression compared to their monolingual peers. This may reflect the bilingual experience of regularly suppressing their second language when not in use, thus facilitating neural efficiency. These findings contribute to the growing body of literature on bilingual cognitive development suggesting that functional connectivity during executive function may differ in bilingual children, even at a young age, despite no observable behavioral differences. This highlights the importance of integrating neuroimaging with behavioral data to gain a more comprehensive understanding of bilingual cognitive development.

Introduction

Executive Function (EF) refers to a set of cognitive processes that enable goal-directed behavior and are primarily supported by the prefrontal cortex (Jones and Graff-Radford, 2021). Inhibitory control falls under EF and refers to the ability to suppress irrelevant thoughts or actions (Rothbart and Posner, 1985). Inhibitory control undergoes rapid development during early childhood (3–5 years old, Moriguchi, 2014, Zelazo and Carlson, 2012) and is crucial for later academic achievement, emotion regulation, and social success (Anthony and Ogg, 2020; Beisly and Jeon, 2024; Sasser et al., 2015). A longitudinal study further supports the importance of inhibitory control over the lifespan, finding that high self-control skills between ages 3 and 11 resulted in significantly better health, higher incomes, and lower rates of criminal conviction in adulthood (Moffitt et al., 2011).

Bilingualism has been proposed as one factor that may influence the development of EF through neural adaptation. Bialystok (2024) proposes that bilingualism facilitates “adaptation,” where neural networks are modified through experience, potentially enhancing attention skills across multiple domains. Exposure to multiple languages early in development may improve attentional control by requiring children to listen to and process multiple linguistic inputs (Bialystok and Craik, 2022). This enhanced attention control may give bilingual children an advantage on complex tasks that demand high levels of sustained attention (Bialystok, 2024). Despite being related constructs, findings of a bilingual advantage have typically been shown with tasks that require interference suppression as opposed to response inhibition (e.g., Simon and Stroop tasks; Martin-Rhee and Bialystok, 2008; Nayak et al., 2020). However, while some studies support a bilingual advantage in interference suppression, findings remain mixed, with almost twice as many studies documenting an advantage in older children (ages 6–12) than in younger children (Hilchey and Klein, 2011; Planckaert et al., 2023).

These inconsistencies in behavioral findings have led some researchers to suggest that cultural and socioeconomic status (SES) factors such as education, occupation, and income may confound the results (Paap, 2022). A meta-analytic review revealed only a small effect of bilingualism on EF, which disappeared once they adjusted for publication bias (Lowe et al., 2021). However, a Bayesian inference analysis found that bilingual children outperform monolinguals in EF “far more often than chance,” even when controlling for publication bias, year, sample size, and task type (Yurtsever et al., 2023). These conflicting findings highlight the need to investigate bilingualism’s effect using neuroimaging techniques rather than relying solely on behavioral measures (Pliatsikas, 2024). The use of neuroimaging techniques can provide additional information on the underlying cognitive functions that may not be adequately captured by behavioral studies (Morita et al., 2016).

Neuroscience research suggests that as inhibitory control matures, neural activation becomes more specialized and efficient, shifting from broad, global activation to a more localized pattern in the prefrontal cortex (Fiske and Holmboe, 2019; Hwang et al., 2010). This aligns with the neural efficiency hypothesis, which posits that greater proficiency in cognitive tasks becomes more automated by adapting to the demands, thus requiring only specific brain regions and less global activation (Debarnot et al., 2014). This further aligns with the principle of efficient coding which states that the brain works to transmit the maximum amount of information in the way that is most metabolically efficient, leading to reduced neural load for behavioral performance (Zhou et al., 2022). Given this framework, examining brain activity can provide deeper insight into how bilingualism influences inhibitory control at the neural level.

Evidence suggests that bilinguals may develop more efficient attentional control, particularly in tasks requiring interference suppression, where irrelevant information must be ignored. For example, in a response inhibition task, Mehnert et al. (2013) found increased connectivity for children ages 4–6 in the bilateral frontal and parietal cortices during both response and inhibition trials using functional near-infrared neuroimaging (fNIRS). Conversely, the adult participants activated only during the inhibition trials and only within the right frontal and parietal cortex. Their results align with the neural efficiency hypothesis, suggesting that younger children, who have less practice with inhibition, recruit neural networks that are broader and not fully specialized compared with adults who have had more practice and are more proficient. Bilingualism, by providing additional practice in attentional control, may accelerate this neural adaption process, leading to less global activation and greater efficiency during inhibition, even in young children (Li et al., 2023).

Bilingualism has been associated with structural brain changes in both grey and white matter, particularly as second language proficiency increases (García-Pentón et al., 2016; Luk and Pliatsikas, 2016; Pliatsikas, 2020). These structural changes in the brain have been shown to influence functional connectivity patterns (Wang and Tao, 2024) and behavioral outcomes for cognitive skills (Gavett et al., 2018). This may, in turn, contribute to a potential bilingual advantage or more efficient cognitive processes in childhood. Task-based functional connectivity measured with fNIRS assesses how different brain regions interact and coordinate their activity during the performance of a specific task. It reflects the temporal correlation in oxygenated hemoglobin fluctuations between regions, revealing which areas co-activate during task engagement and thus may be functionally linked. This approach helps identify networks of coordinated neural activity that support specific cognitive processes, such as inhibitory control (Linnman et al., 2012; Chong et al., 2025). In the context of inhibitory control, studies examining interference suppression have found increased neural activation in areas of the prefrontal cortex, including the bilateral dorsolateral-prefrontal cortex [dlPFC; Brodmann areas (BAs) 9 and 46], medial prefrontal cortex (mPFC; BA 10) and bilateral inferior frontal gyrus (IFG; BA 45; Peters and Smith, 2020; Tao et al., 2021; Wang and Tao, 2024). These findings highlight the importance of investigating how bilingualism may shape functional connectivity within these networks during interference suppression tasks.

Despite these advances, understanding the functional impact of bilingualism on interference suppression remains challenging, particularly in young children. However, a recent review highlights fNIRS as a promising tool for refining existing theories on the effects of bilingualism and cognition (see Pliatsikas, 2024 for a review). Importantly, the review notes that fNIRS studies have yet to provide consistent evidence on how bilingualism influences brain function, underscoring the need for further research to clarify these effects. For instance, a study found no association between second language on functional brain activation during a task (Moriguchi and Lertladaluck, 2020), whereas other studies have found those positive associations (Arredondo et al., 2022; Xie et al., 2021).

Furthermore, a recent fNIRS study found evidence that suggests that bilingualism may promote neural efficiency in the PFC. It found that bilingual children required fewer cortical resources during a card sort interference/switching task, suggesting that bilinguals require fewer neural resources to achieve similar behavioral performance as monolinguals (Li et al., 2023). Building on these findings, the current study leverages fNIRS to examine functional connectivity within the bilateral dlPFC, mPFC, and bilateral IFG during interference suppression task. This will provide deeper insights into how bilingual and monolingual preschoolers differ in their neural processing of interference suppression.

By incorporating fNIRS alongside behavioral measures of interference suppression in early childhood, the current study aims to enhance our understanding of the bilingual experience of young children during inhibition. Based on conceptualizations of bilingualism promoting attentional control (Bialystok, 2024; Bialystok and Craik, 2022), we hypothesize that bilingual preschoolers will demonstrate similar behavioral performance on the congruent trials requiring less attentional control and better performance on the incongruent trials of a Simon-like interference suppression task than age-matched monolingual preschoolers. We further hypothesize task-state functional connectivity region of interest pattern differences within the prefrontal cortex with bilingual preschoolers, demonstrating more efficient neural processing (i.e., less global connections) than monolingual preschoolers.

Materials and methodsParticipants

The current study utilized data from 26 preschool aged children recruited from an on-campus childcare from a large public university in the United States of America and the surrounding community. Parents of participating children were asked to complete online questionnaires describing family characteristics and home language use. The current sample included 13 bilingual children (female = 9, mean age = 59.39 months) whose home languages were primarily of East Asian or Ibero-Romantic descent and 13 age matched monolingual English children (female = 5, mean age = 59.22 months). Most participating parents for bilingual children had a bachelor’s degree (n = 5) followed by those with a doctorate (n = 4) then a master’s degree (n = 3). One parent had completed some college. Most participating parents for monolingual children had a bachelor’s degree (n = 7) followed by a doctorate (n = 2), master’s degree (n = 2), and having completed some college (n = 2). Second languages spoken at home include Mandarin (n = 4), Portuguese (n = 3), Spanish (n = 3), Korean (n = 1), German (n = 1), and French (n = 1). Descriptive statistics for all continuous and categorical variables are presented in Table 1.

CharacteristicMonolingualBilingualTotal sampleMeanSDMeanSDMeanSDAge (months)59.229.2759.399.2759.309.08English proficiency (PPVT)128.717.64115.223.23121.9521.24Effortful control68.256.9265.6913.1966.9210.52n%n%n%SexFemale5389691454Male8624311246Parent educationSome college21518312Bachelors7545381246Masters215323519Doctorate215431623Second languageMandarin431Portuguese323Spanish323Korean18German18French18Pairwise comparisonCongruent RTIncongruent RTt(12)p-valueM (ms)SD (ms)M (ms)SD (ms)Monolingual1474.06285.961581.07370.31−2.120.055Bilingual1403.94323.731531.85413.55−1.240.239

Demographic characteristics.

N = 26 (n = 13 for each group).

Bilingualism can be a complex construct, with definitions varying based on age and study purpose. For example, very young children may be considered bilingual if they receive an appropriate amount of exposure to a second language whereas more weight would be given to the level of fluency in older youth and adults. Due to the wide variety of language pairings present in our sample, we selected our bilingual sample based on exposure to a second language rather than proficiency (Loe and Feldman, 2016; Valicenti-McDermott et al., 2013; Singh et al., 2015). For this study, bilingualism is defined as having at least 25% exposure to a second language and monolingualism is defined as having at least 90% exposure to a first language (Pearson et al., 1993). The present study utilized a language background questionnaire (LBQ) adapted from a phone-based questionnaire (Singh et al., 2015) where parents were asked about the different home languages to which their children were exposed. Bilingual participants in our sample had an average of 74% (45–100%) second language exposure from parents. Monolingual children were marked as zero exposure to a second language. All parents of participants signed an IRB approved consent form and received monetary compensation for participating. Children provided assent and received a book or toy of their choosing.

MeasuresPeabody picture vocabulary test—fourth edition

English receptive vocabulary was measured using the Peabody Picture Vocabulary Test—Fourth Edition (PPVT-4) (Dunn and Dunn, 2007) to ensure that all of the children, regardless of their home language, had similar levels of understanding in English. The PPVT-4 is a norm-referenced task that is individually administered by a trained researcher. The task requires the child to choose the picture that best represents the stated word. The child has four picture options to choose from and the task ends when the child misses 8 or more words in a set of 12. The PPVT-4 provides a standardized score based on age and sex.

Child’s behavior questionnaire—effortful control

Similarity between monolingual and bilingual parents’ perceptions of various aspects of their children’s behaviors (e.g., inhibitory control, attention, and emotion regulation), was measured using the effortful control subscale of the child’s behavior questionnaire very short form Putnam and Rothbart (2006) and Rothbart et al. (2001). Parents answered 12 questions related to child effortful control (α = 0.88; e.g., Is good at following instructions.) and a sum score was computed. This measure was included to examine whether caregivers’ reports of the aforementioned behaviors differed between bilingual and monolingual preschoolers and to provide an additional context beyond our laboratory setting.

Spatial conflict arrows (arrows)

Interference suppression was assessed using the Spatial Conflict Arrows (Arrows; Figure 1) task from the EF Touch battery, a computerized set of tasks developed for children aged 3–5 years (Willoughby et al., 2012a; Willoughby et al., 2012b). This Simon-like task (Simon, 1969) measures interference suppression by requiring children to respond to the direction an arrow is pointing, regardless of its location on the screen. In congruent trials, the arrow appears on the same side it points to (e.g., pointing left on the left side), while in incongruent trials, the arrow appears on the opposite side (e.g., pointing left on the right side). All participants completed the same sequence of 36 trials (19 congruent, 17 incongruent), presented in a single block. The task began with 12 congruent trials, followed by 12 incongruent trials, and ended with a mixed sequence. For the first six participants (split evenly between monolingual and bilingual), trials were presented for 2 s, which was later adjusted to 4 s for the remaining participants to better accommodate younger children.

Congruent and incongruent trials for spatial conflict arrows. (A) Represents congruent trials and (B) represents incongruent trials. Children were required to respond based on the direction the arrow is pointing and not the location of the arrow.

Accuracy was scored as the proportion of correct responses (0–1.00), and reaction time was recorded in milliseconds. Performance was analyzed separately for congruent and incongruent trials. Group comparisons between bilingual and monolingual children were conducted using Mann–Whitney U tests for accuracy (based on non-normal distributions) and independent-samples t-tests for reaction time. Normality was assessed using the Shapiro–Wilk test, appropriate for small sample sizes (N < 50) (Ghasemi and Zahediasl, 2012). All statistical analyses were performed using R version 4.4.2 (R Core Team, 2024).

fNIRS data acquisition and processing

fNIRS data were collected in a secured room using a continuous-wave NIRSport2 system (NIRx, Medical Technologies, LLC). A NIRx prefrontal cortex montage containing eight light sources that emit near-infrared light at wavelengths of 760 and 850 nm and seven detectors (Figure 2) was used. This resulted in 20 channels of data being collected. Prior to the neuroimaging session, parents and children provided consent/assent, and efforts were made to ensure participants’ comfort throughout. Prior to data collection, participants were fitted with the cap which was then calibrated to get the best possible signal. If channels had poor connections, the cap was readjusted, and hair was parted to allow for sources and detectors to have more direct contact with the scalp.

Prefrontal 8 × 8 cap montage. Red circles are source locations and blue circles are detector locations. Location labels follow the EEG 10/10 system. Researchers received verbal consent from the mother and assent from the child to take this photo. Maternal consent was also received to include a deidentified copy of the photo for publication.

Data were exported from the NIRSport2 device into MATLAB version R2022b (The MathWorks Inc., 2022) where the NIRS Brain AnalyzIR Toolbox (Santosa et al., 2018) was used to preprocess the fNIRS data, extract the oxygenated (HbO) and deoxygenated (HbR) hemoglobin values, map the channel coordinates onto the Brodmann areas of interest (BAs 8, 9, 10, 45, and 46) and calculate the inter- and intra-individual correlations between fNIRS channels for both HbO and HbR. The current analyses focused on changes in hemoglobin concentration and functional connectivity for only HbO values as they are more sensitive to task related changes and have a higher signal amplitude than HbR values (Hoshi, 2003; Luke et al., 2021). Prior to preprocessing the fNIRS data, event triggers for all trials were manually input based off an initial task-start trigger from data collection. Trigger timings were calculated following the previously mentioned pattern of trials. These markers were then used for both analyses. First, raw data were visually inspected for noisy channels. Following the visual inspection, raw values were then converted to optical density values and then a temporal derivative distribution repair (Fishburn et al., 2019) was applied to help correct motion artifacts. Next, optical density values were then converted to HbO and HbR values via the modified Beer–Lambert Law (Jacques, 2013). For the functional connectivity analysis, converted data was down sampled from 10 to 1 Hz in order to reduce the impact of serial autocorrelations on analyses (Pinti et al., 2019). However, for the region of interest analysis, data was not down sampled. Finally, due to the unique statistical properties of fNIRS, an autoregressive, iteratively reweighted robust model (pre-whitening) was used to improve model robustness and account for serial dependencies in the data (Barker et al., 2013; Huppert, 2016).

Region of interest analysis

Participant neural activity (beta values) were calculated by solving a generalized linear model (GLM) for each channel within every participant for both trial types. Level one of the GLM included individual HbO values during task completion. Level-one analysis also included regressors to model channel activation during each task condition regardless of group. Level-two of the GLM explored main effects of condition for each group. Additionally, between group comparisons for both condition types were examined at this level. The formula for the level-two GLM is displayed here [beta ~ − 1 + group:cond + (1|ID)].

Connectivity analysis

Individual level bivariate correlations were calculated between each channel pairing to determine task-state functional connectivity across the prefrontal cortex. The first 24s1 for each trial type (i.e., congruent and incongruent) were utilized for the connectivity analyses. Analyzing trial types individually allows us to capture the nuanced activation patterns and gives us a better overall picture of neural efficiency. Prior to the connectivity analyses, channel coordinates were registered to the Colin 27 atlas with a 53 cm circumference. Task-state functional connectivity was assessed using auto-regressive whitened robust correlations between channel pairs, for HbO. Individual-level connectivity patterns were first calculated using the “connectivity” function from the Nirs Brain AnalyzIR toolbox (Santosa et al., 2018). This function calculates an all-to-all connectivity matrix via an autoregressive, robust correlation with a maximum model order of 4x the sampling rate. Group level models were then calculated using the “MixedEffectsConnectivity” function. The formula for the mixed effects model is presented here (R ~ −1 + group:cond) (for additional information on the mathematical procedures and functions see Barker et al., 2013; Lanka et al., 2022; Santosa et al., 2017; Santosa et al., 2018). To determine significance of channel correlations, one-sample t-tests were conducted and a False Discovery Rate (FDR) correction (Benjamini and Hochberg, 1995) was used to account for multiple comparisons.

ResultsDescriptives

Children in the bilingual and monolingual groups did not differ in age [t(26) = −0.04, p = 0.97], parent report of effortful control (U = 86.50, z = 0.46, p = 0.663, r = 0.09) or receptive English language score as measured by the PPVT-4 [t(16.8) = 1.46, p = 0.162]. Additionally, bilingual and monolingual groups did not differ in terms of parental education levels [t(23.9) = −0.45, p = 0.26]. These findings suggests that both bilingual and monolingual children were comparable in age, English receptive language proficiency, parent reported effortful control behavioral scores, and parental education levels, despite differences in language exposure.

Interference suppression behavioral dataWithin group differences

To test for a within group interference effect between correct responses for congruent and incongruent trials of the spatial conflict arrows task, paired samples t-tests were conducted for monolingual and bilingual preschoolers. For the bilingual preschoolers, the results were non-significant [t(12) = −1.24, p = 0.24] suggesting that the bilingual preschoolers did not experience an interference effect during this task. For the monolingual preschoolers, the results were also non-significant [t(12) = −2.12, p = 0.055], however, as the p-value was 0.055 and was approaching significance, this may indicate that there is potentially an interference effect between trial types for monolingual preschoolers and should be tested further with a larger sample.

Between group differences

To test if there are group differences in task performance during the spatial conflict arrows task, Mann–Whitney U tests were conducted due to non-normality in the dependent variables based on the results of the Shapiro–Wilk test which is recommended for small sample sizes (Ghasemi and Zahediasl, 2012; Shapiro et al., 1968; WCong = 0.832, p < 0.001; WInong = 0.917, p = 0.038; WCombined = 0.902, p = 0.017). Accuracy comparisons between bilingual (MdnCong = 0.89, IQR = 0.16; MdnIncong = 0.71, IQR = 0.18; MdnCombined = 0.81, IQR = 0.19) and monolingual (MdnCong = 0.84, IQR = 0.21; MdnIncong = 0.88, IQR = 0.29; MdnCombined = 0.86, IQR = 0.28) preschoolers for task accuracy split by trial type and combined are presented in Table 2. The results indicated that there was not a significant difference in the accuracy between groups (UCong = 84, z = −0.03, p = 0.999, r = 0.01; UIncong = 111.50, z = 1.39, p = 0.171, r = 0.27; UCombined = 102.50, z = 0.93, p = 0.368, r = 0.18) for congruent and incongruent tasks. To test if there are group differences between bilingual (MCong = 1403.94 ms, SD = 323.73 ms; MIncong = 1531.85 ms, SD = 413.55 ms; MCombined = 1415.27 ms, SD = 343.51 ms) and monolingual (MCong = 1474.06 ms, SD = 285.96 ms; MIncong = 1581.07 ms, SD = 370.31 ms; MCombined = 1484.10 ms, SD = 374.36 ms) preschoolers in task reaction time Welch’s two-sample t-tests were conducted as reaction time was normally distributed (WCong = 0.944, p = 0.169; WIncong = 0.981, p = 0.888; WCombined = 0.926, p = 0.061). Reaction time comparisons between groups are presented in Table 3. The results for the t-test indicated that there was also not a significant difference between groups [tCong(23.6) = 0.59, p = 0.564, d = 0.23; tIncong(23.7) = 0.32, p = 0.752, d = 0.13; tCombined(23.8) = 0.49, p = 0.630, d = 0.19] for the congruent and incongruent tasks. This Suggests that both groups were able to respond similarly on both congruent and incongruent tasks of the Simon-like task measuring interference suppression.

Trial typeMonolingual (n = 13)Bilingual (n = 13)U(24)Z-scorep-valueMdnIQRMdnIQRCongruent accuracy0.840.210.890.1684−0.03>0.999Incongruent accuracy0.880.290.710.18111.501.390.171Combined accuracy0.860.220.810.19102.500.930.368

Group comparisons of response accuracy for congruent and incongruent trials.

Groups were compared via Mann–Whitney U tests.

Trial typeMonolingual (n = 13)Bilingual (n = 13)t(23.6)Cohen’s dp-valueMSDMSDCongruent RT1474.06285.961403.94323.730.590.230.564Incongruent RT1581.07370.311531.85413.550.320.130.752Combined RT1484.10374.361415.27343.510.490.190.630

Group comparisons of reaction time for correct congruent and incongruent trials.

Groups were compared via Welch’s Two Sample t-test. Reaction times are in milliseconds.

Region of interest analysis results

To evaluate how bilingual and monolingual preschoolers differ in patterns of neural recruitment during interference suppression, a generalized linear model analysis was conducted. For congruent trials, neither monolingual or bilingual participants had significant activation for any channels at an FDR corrected q-value of 0.05. For incongruent trials, only the monolingual participants had significant activation in channels S1-D2, β = 12.49, SE = 3.46, t(46) = 3.61, q = 0.04, and S4-D5, β = 16.25, SE = 3.39, t(46) = 4.79, q < 0.01.

For congruent trials, there were no significant differences in channel activation at a q < 0.05 between monolingual and bilingual preschoolers. Channels S1-D2, β = 12.49, SE = 3.46, t(46) = 3.61, q = 0.01, and S4-D5, β = 16.25, SE = 3.39, t(46) = 4.79, q < 0.001.

A bar plot for fNIRS channels averaged into regions of interest based on the channel weights (