Neural Correlates of Morphosyntactic Processing in Spanish–English Bilingual Children: A Functional Near-Infrared Spectroscopy Study

Abstract

Purpose:

The aim of this study was to examine the effects of early bilingual exposure on Spanish–English bilingual children's neural organization of English morphosyntactic structures. This study examines how children's age and language experiences are related to morphosyntactic processing at the neural level.

Method:

Eighty-one children (ages 6–11 years) completed an auditory sentence judgment task during functional near-infrared spectroscopy neuroimaging. The measure tapped into children's processing of early-acquired (present progressive –ing) and later-acquired (past tense –ed and third-person singular –s) English morphosyntactic structures, the primary language of academic instruction.

Results:

We observed effects of syntactic structure and age. Early-acquired morphemic structures elicited activation in the left inferior frontal gyrus, while the later-acquired structures elicited additional activations in the left middle temporal gyrus and superior temporal gyrus (STG). Younger children had a more distributed neural response, whereas older children had a more focal neural response. Finally, there was a trending association between children's English language use and left STG activation for later-acquired structures.

Conclusion:

The findings inform theories of language and brain development by highlighting the mechanisms by which age and language experiences influence bilingual children's neural architecture for morphosyntactic processing.

Neurobiological foundations for language are jointly shaped by children's everyday experiences and brain maturation. Bilingualism is one of the most common forms of life experiences that profoundly influences neurodevelopmental processes (Hernandez et al., 2019; Werker & Hensch, 2015). However, the effects of childhood bilingualism on the neurobiology of language functions remain largely unexplored. Models of brain development for language have been put forth yet are primarily built on monolingual evidence and generally exclude the widespread bilingual acquisition phenomena. It is vital for us to complement and expand on this knowledge as more than half of the world's population speak two or more languages. Morphosyntactic knowledge is one such area that requires attention, as it is a strong predictor of children's language comprehension, language production, and reading skills—competencies commonly used as benchmarks for academic achievement. Thus, this study examines the effects of Spanish–English bilingualism on children's neural organization for morphosyntactic processing. Using functional near-infrared spectroscopy (fNIRS), we assess how age and language use influence the engagement of bilingual children's language network during English sentence comprehension.

In monolingual English-speaking children, present progressive –ing is typically the first grammatical morpheme children acquire around 27–30 months of age (Brown, 1973; de Villiers & de Villiers, 1973). Third-person singular –s (e.g., “He walks”) and regular past tense –ed (e.g., “He walked”) are acquired later, around 35–40 months of age (Brown, 1973). Preschoolers have been found to display different event-related potential patterns to syntactic (–ing omission) violations within sentences, showing that children even as young as 3 years of age exhibit neural recognition responses to morphosyntactic structures (Silva-Pereyra et al., 2005). Around 6 years of age, children show adultlike accuracy on grammatical judgments of finiteness, tense, and agreement (Rice et al., 1995). Early school–age children with a developmental language disorder (DLD) tend to exhibit morphosyntactic errors, rather than those of semantic depth and breadth. Errors in the production of past tense –ed and third-person singular –s are hallmark errors of DLD in early elementary school grades (e.g., Marchman et al., 1999; Norbury et al., 2001; Rice et al., 1998, 2004). By examining the neural organization of morphosyntax in neurotypical Spanish–English bilingual children, we can provide a foundation to understand differences noted behaviorally in bilingual children with DLD.

Morphosyntactic development is language specific, and in bilinguals, languages are functionally independent. However, there are cross-linguistic influences or transfer effects between the languages, leading to differences in language acquisition between monolingual and bilingual speakers (e.g., Baron et al., 2018; Bedore & Peña, 2008; Kohnert, 2010). Additionally, there is a high level of heterogeneity between bilingual children due to the age of acquisition, the amount they hear and speak each language at any given time, proficiency level, and type of linguistic task (e.g., Paradis et al., 2011). To distinguish between the influence of age of acquisition and proficiency level, researchers have noted that lexical-semantic processing is more dependent on proficiency level while syntactic processing is more influenced by the age of acquisition (e.g., Hernandez & Li, 2007; Wartenburger et al., 2003).

Behaviorally, the amount of exposure to English has a significant impact on morphosyntactic acquisition in bilinguals. Davison and Hammer (2012) conducted a study with preschool-age bilingual children who differed in their exposure to English. All were exposed to Spanish from birth, but one group was exposed to English before starting preschool at 3 years of age while the other had very minimal, if any, exposure to English. After 2 years in preschool, children with early English language exposure mastered present progressive –ing early, three fourths of the children mastered past tense –ed, and third-person singular –s was acquired by only a quarter of the children. For the children with later English language exposure, three fourths of the children mastered present progressive –ing, half mastered past tense –ed, and very few children mastered third-person singular –s. Thus, it is notable that bilingual preschool children have significant variability in the mastery of their English grammatical morphemes and their patterns of development differ from their monolingual peers. In another study focusing on bilingual Spanish–English prekindergarten and kindergarten children, Bohman et al. (2010) concluded that the amount children heard of each language day-to-day (exposure) strongly influenced children's semantic knowledge while morphosyntax knowledge was influenced by both exposure and use (amount the child says). These studies point to the specific role that exposure and use play in morphosyntactic processing and development.

The age of acquisition and language use differentially influence the engagement of the language network in bilinguals as well (e.g., Cargnelutti et al., 2019; Jasinska & Petitto, 2013; Pallier et al., 2003; Perani et al., 1998; Sulpizio et al., 2020; Sun et al., 2023; van Heuven & Dijkstra, 2010). Neurotypical monolingual adults often engage the left inferior frontal cortex, temporal, and inferior parietal regions during sentence processing tasks (for reviews, see Hagoort, 2019; Hagoort & Indefrey, 2014). Theoretical frameworks of language development pose that brain specialization for language proceeds along the caudorostral trajectory such that the temporal regions specialize for language first, supporting the encoding of core linguistic representations, whereas the specialization of the frontal regions follows a more protracted trajectory, which developmentally supports more efficient access to the representations (e.g., Enge et al., 2020; Skeide et al., 2014, 2016). In a recent meta-analysis of functional magnetic resonance imaging (fMRI) language comprehension research, Enge et al. (2020) reported on over 600 monolingual children across 24 studies. Compared to adults, children's functional activation was more widespread that included distributed bilateral temporal gyri and left inferior frontal gyrus (IFG) areas including pars triangularis and pars orbitalis. Adults showed more focal and reduced activation in the left superior temporal gyrus (STG) and left IFG pars opercularis. Individual studies with monolingual learners suggest that adultlike functional patterns may emerge around ages 9–10 years, during which there is a shift toward focal and specialized left IFG engagement for syntactic processing (e.g., Brauer & Friederici, 2007; Nuñez et al., 2011; Skeide et al., 2014; Wang et al., 2020, 2021; Wu et al., 2016).

Much of the bilingual neuroimaging literature has focused on adult bilinguals; however, there are a few studies that have investigated English sentence processing in bilingual children using fNIRS. One study focused on English monolingual children and early- and later-exposed bilingual children (ages 7–10 years) during plausible and implausible sentences in object–subject or subject–object order (Jasinska & Petitto, 2013). All children had been exposed to English from birth and learned a second language early (ages 0–3 years) or later (ages 4–6 years). Greater neural activation was observed for later-exposed bilinguals when compared with early-exposed bilinguals in bilateral STG and dorsolateral prefrontal cortex (DLPFC) as well as the right inferior parietal lobule. Later-exposed bilinguals showed greater neural activation when compared to monolingual children in bilateral STG and DLPFC. In summary, the age of first exposure to a second language influences the functional engagement of the brain when processing the first language. In a recent study by Arredondo et al. (2019) on monolingual and Spanish–English bilingual children (ages 6–12 years), all children showed greater neural activation in the left IFG for English sentences with violations in later-acquired (tense/agreement) structures than earlier-acquired grammatical structures (–ing). Bilingual children demonstrated stronger and more restricted activation in the left IFG (Arredondo et al., 2019). Similarly, Wagley et al. (2023) examined Spanish–English bilingual children (ages 7–12 years) using the same sentence judgment task as this study. Bilingual children showed significant activation in the left IFG for English sentences with violations in later-acquired structures (–ed and –s) when compared to grammatically correct sentences. Importantly, activation in the pars opercularis regions of the left IFG was significantly related to children's behavioral morphosyntactic knowledge in Spanish. Together, results from these studies suggest a link between early bilingual experiences and the functional activations for language, highlighting cross-linguistic influences for morphosyntactic processing in the brain.

The Present Study

This study examines the influence of dual language experiences on Spanish–English bilingual children's brain organization of early-acquired (present progressive –ing) and later-acquired (past tense –ed and third-person singular –s) English morphosyntactic structures. Additionally, this work further looks at how morphosyntactic processing is related to age and current English language use. Our neuroimaging task specifically focused on English as it is the primary language of children's academic instruction across the United States.

We utilize fNIRS technology to measure changes in the brain tissue (i.e., functional activation) when processing linguistic information. fNIRS monitors changes in optical properties as an indicator of neuronal activity (e.g., Logothetis, 2002). For this purpose, near-infrared light (650–1000 nm) is employed where light propagates into scattering tissues and its absorption by oxygenated and deoxygenated hemoglobin (HbO and HbR; e.g., Cutini & Brigadoi, 2014). NIRS signals are highly correlated with regional cerebral blood flow and resemble positron emission tomography and fMRI measurements (e.g., Huppert et al., 2006). Additionally, fNIRS puts fewer constraints on natural movements (e.g., head tilts), is smaller in size and more portable to a variety of research settings, and is quiet as compared to other brain imaging methodologies. Thus, fNIRS has been gaining traction in studying pediatric and bilingual populations (Arredondo, 2023; Nickerson & Kovelman, 2022). Using an experimental task specifically designed for fNIRS, as well as behavioral and survey data, this study addresses the following questions in 81 Spanish–English bilingual children ages 6–11 years:

1.  What are the neural correlates of early- and later-acquired acquired English morphosyntactic structures in Spanish–English bilingual children?

For Research Question 1, we hypothesized that the left IFG, left STG, and left middle temporal gyrus (MTG) would be primarily recruited for morphosyntactic processing during sentence comprehension. Based on previous literature (Friederici, 2009; Friederici et al., 2017; Skeide, 2012; Wang et al., 2021; Xiao et al., 2016), we expected to see the strongest activation (HbO signal change) within the left IFG pars opercularis and triangularis and posterior half of the left STG/MTG.

2.  Does functional activation for morphosyntactic processing vary with age (maturational state of the brain)?

For Research Question 2, we hypothesized that older children would have a greater and more restricted (focal) activation in the left IFG for earlier-acquired (present progressive; ING) and later-acquired (past tense and third-person singular; ED&S) morphosyntactic structures than younger children, as their cortical response should be more adultlike. Younger children would have more widely spread (distributed) activation for later-acquired structures as they would not have as well-developed functional specialization for syntactic processing (e.g., Knoll et al., 2012; Luke et al., 2002; Skeide et al., 2014, 2016; Skeide & Friederici, 2016; Wang et al., 2020, 2021). We additionally hypothesized that there would be no differences detected in the activation of the left STG/MTG between older and younger children on either early- or later-acquired structures. Regarding brain maturity, temporal regions develop earlier than the frontal regions. By ages 5–6 years, the temporal lobe is already showing specificity for morphosyntactic features (Enge et al., 2020; Friederici, 2006; Skeide et al., 2014; Wang et al., 2020; Wang, Wagley, et al., 2022). Thus, by 6 years of age, we do not expect differences between age groups in the left STG.

3.  Does functional activation for morphosyntactic processing vary as a function of current English exposure and use?

For Research Question 3, we hypothesized that bilingual children with more English exposure and use would have more restricted neural activity in both the left IFG and left STG/MTG for earlier- and later-acquired structures than children who have less English exposure and use. Previous neuroimaging literature has focused on age of acquisition and has found differences between early and late bilinguals (e.g., Cargnelutti et al., 2019; Jasinska & Petitto, 2013; Wartenburger et al., 2003). However, within the behavioral literature, there are additional language measures used to explain language performance, including current language exposure and use. For bilinguals, the amount of exposure and experience in each language accounts for the rate of language acquisition and a significant amount of the variance in a child's overall language performance (e.g., Hammer et al., 2012; Paradis et al., 2011; Serratrice & De Cat, 2020). For young bilingual children, grammatical knowledge in each language has been associated with the amount of language experience that they have (Place & Hoff, 2011). Additionally, when measures of both age of acquisition and current language use are available, current use tends to emerge as the stronger predictor of language skill (Ashkenazi et al., 2020; Bedore et al., 2012, 2016). Thus, the focus here is on current language exposure and use to examine if this measure can explain functional activation in the brain for morphosyntactic processing.

Method

The study hypotheses and analytical plan were preregistered through Open Science Framework (OSF) after data cleaning but prior to beginning the data analyses (see https://osf.io/9jmnb). The data used in this study were collected as part of a larger data set assessing language and literacy development in Spanish–English bilingual children (Wagley, 2019; Wagley et al., 2022). The complete data collection session lasted around 4 hr, and participants were compensated $30 and a gift bag with a toy for participation. A subsample of assessments and neuroimaging data were analyzed for this study. This study was approved by the institutional review board at the University of Michigan.

Participants

One hundred thirty-three participants were recruited from communities around southeast Michigan using advertisements posted at schools, community organizations, and on Facebook. Only participants who met the following criteria were recruited for the study: (1) ages 6;0–11;11 (years;months); (2) Spanish–English bilingual with (a) exposure to Spanish at birth and to English prior to 5 years of age, (b) a minimum of two continuous years of daily English use in the United States prior to participation, and (c) at least one native Spanish-speaking parent who reported consistent use of Spanish at home; (3) right-handed; (4) no clinical diagnosis of attention-deficit/hyperactivity disorder or neurological, psychiatric, or developmental disorders as reported in the parent questionnaire; and (5) normal hearing and normal/corrected vision as reported in the parent questionnaire. Twenty-five participants were excluded from the analyses for not meeting the above criteria.

Additionally, inclusion criteria for this study were the following: (a) complete behavioral morphosyntax data on the Bilingual English–Spanish Assessment–Middle Extension (BESA-ME; Peña et al., 2016) in English and Spanish (n = 4 excluded), (b) complete fNIRS imaging data of the experimental grammaticality judgment task (n = 21 excluded), (c) a standard score of above 85 on the Woodcock-Johnson/Muñoz Passage Comprehension subtest in English and Spanish (n = 4 excluded), (d) a standard score of above 80 on the BESA-ME Morphosyntax subtest in English and Spanish (n = 3 excluded), and (e) a response bias less than 50% accuracy difference between the condition –ed and –s inflectional morpheme omission and the correct sentences (n = 6 excluded). All participants had greater than 50% accuracy on the grammatically correct condition of the fNIRS morphosyntax task.

Based on the criteria outlined above, 81 participants were included in the final analyses. Participants were grouped by age into a younger children's group (ages 6;0–8;11, n = 40) and an older children's group (ages 9;0–11;11, n = 41). In the preregistration, we reported that 42 participants were included in the older children's group. However, one participant was excluded during analyses as the participant did not meet all inclusionary criteria.

Procedure

Prior to testing, parents completed a background questionnaire over the phone to determine their child's eligibility, along with the language history questionnaire (see details below). Eligible participants were invited for a lab visit. During the lab visit, parents completed an additional background and demographics questionnaire (e.g., Kovelman et al., 2008), while children completed language and literacy standardized assessments in English and Spanish and an experimental task during fNIRS imaging. Testing was completed in one session.

Language History Questionnaire

To examine a child's everyday bilingual language exposure and use, parents completed the Bilingual Input/Output Survey (Peña et al., 2018) describing the quantity of their child's language use to the best of their ability. This questionnaire asked parents to detail a typical weekday and a typical weekend day of the child on an hour-by-hour basis, including the language(s) the child is exposed to inside and outside the home. Based on this hour-by-hour report, we calculated the number of hours children spent hearing (input) and speaking (output) each of their languages and a relative percentage of time spent using each language.

Standardized Behavioral Assessments

Children participated in behavioral assessments of language and literacy in Spanish and English. Children's language comprehension skills were measured using the Semantic Knowledge and Morphosyntax Knowledge subtests of the BESA-ME. This assessment examines knowledge of grammatical morphemes and sentence structure, more broadly, using cloze task and sentence repetition items in each language. Children's reading comprehension skills, used here as an inclusionary criterion, were assessed using the Passage Comprehension subtest of the Woodcock-Johnson III Normative Update (Tests of Achievement; Woodcock et al., 2001) and Batería III Woodcock-Muñoz Normative Update (Pruebas de Aprovechamiento; Woodcock et al., 2004). All participant demographics and information regarding their language abilities are included in Table 1. Task performance and language abilities are reported in Table 1 for the entire group of young and older bilingual children.

Table 1.

Participant demographics, descriptive statistics, and performance on bilingual assessments of language and literacy (standard scores).

Variable	Young (n = 41)		Old (n = 40)
	M	SD	M	SD
Age (years)	7.75	0.83	9.92	0.69
First words in English (years) a	1.23	1.33	1.44	1.25
English input (%)	71.30	9.17	68.30	6.17
English output (%)	72.10	9.72	71.10	8.10
Combined English input/output (%)	71.80	9.34	69.70	6.64
Parent 1 education b	6.10	1.32	5.54	1.55
Parent 2 education b	5.87	1.38	5.78	1.68
Parent 1 occupation c	3.35	3.84	2.66	3.31
Parent 2 occupation c	7.11	2.51	6.98	2.56
Percentage of children who qualified for FRLP	19.51		25.00
Household income d	4.86	1.27	4.41	1.48
BESA-ME English Morphosyntax Knowledge standard score	114.0	17.8	116.0	8.79
BESA-ME Spanish Morphosyntax Knowledge standard score	85.9	24.1	90.6	25.0
BESA-ME English Semantic Knowledge standard score	115.0	12.9	113.0	14.3
BESA-ME Spanish Semantic Knowledge standard score	89.2	15.9	88.3	21.2
WJ English Passage Comprehension***	104.0	9.14	98.1	9.56
WM Spanish Passage Comprehension*	92.9	15.1	84.7	18.0

Note. Group differences using an independent-samples t test between the younger and older groups are denoted with *p < .05 and ***p < .001. FRLP = Free and Reduced Lunch Program; BESA-ME = Bilingual English–Spanish Assessment–Middle Extension; WJ = Woodcock-Johnson; WM = Woodcock-Muñoz.

^a Parents were asked when the participant spoke their first words in English. If first words emerged at 0–1 years of age, they were scored as 0; if words emerged at 2 years of age, they were scored as 1; and so on.

^b Measured using the following scale: 1 = primary and secondary school, 2 = GED and associate degree, 3 = bachelor's degree, 4 = master's and doctoral degree.

^c Measured using the following scale: 0 = not applicable or unknown, students, housewives, unemployed; 1 = farm laborers, menial service workers, students, housewives (dependent on welfare, no regular occupation); 2 = unskilled workers; 3 = machine operators and semiskilled workers; 4 = smaller business owners (< $25,000), skilled manual laborers, craftsmen, tenant farmers; 5 = clerical and sales workers, small farm and business owners (business valued at $25,000–$50,000); 6 = technicians, semiprofessionals, small business owners (business valued at $50,000–$70,000); 7 = smaller business owners, farm owners, managers, minor professionals; 8 = administrators, lesser professionals, proprietor of medium-sized business; 9 = higher executive, proprietor of large businesses, major professional.

^d Measured using the following scale: 0 = < $12,000, 1 = $12,000–$15,999, 2 = $16,000–$49,999, 3 = $50,000–$74,999, 4 = $75,000–$100,000, 5 = > $100,000.

fNIRS Experimental Task

The early-acquired morphosyntactic structure is the present progressive –ing. The later-acquired morphosyntactic structures are regular past tense –ed and third-person singular –s.

Grammaticality was measured using an experimental sentence judgment task developed for fNIRS imaging. Participants heard a sentence in English and indicated whether the sentence was grammatically correct or not. There were three experimental conditions in the task, with 20 sentences in each condition: grammatically correct (CORR), –ing inflectional morpheme omission (ING), and –ed and –s inflectional morpheme omission (ED&S). In the ING condition, children heard a sentence where –ing verb endings were omitted (e.g., “Right now, he is walk_ his dog”). In the ED&S condition, children heard sentences where either the –ed or –s verb endings were omitted (e.g., “Laura score_ a winning goal,” “Yesterday, they finish_ all of the homework”). Ten sentences in the ED&S condition omitted the –s ending (e.g., “Nicholas bite_ into a pizza”), and 10 sentences omitted the –ed ending (e.g., “Last week, they laugh_ with grandma”). In the CORR condition, children heard sentences with grammatically correct verb endings (e.g., “Carmen is tying her shoelaces,” “Yesterday, Daniel picked flowers”). Of the 20 correct sentences, 10 sentences included –ing endings, five sentences included –ed endings, and five sentences included –s verb endings. Overall task performances for each condition are reported in Table 2.

Table 2.

Mean (SD) for accuracy and response time on the grammaticality judgment task.

Variable	Accuracy (%)		Response time (ms)
	Young	Old	Young	Old
Grammatically correct sentences	87 (12)	92 (10)	3,010 (447)	3,081 (565)
ING omission sentences	91 (17)	95 (7)	3,177 (428)	3,165 (475)
ED&S omission sentences	75 (15)	74 (15)	3,225 (409)	3,372 (448)

Note. Reaction times were calculated only based on trials in which the participant responded accurately (i.e., with the correct judgment response).

Sentences were developed based on the grammaticality task described by Wang, Lytle, et al. (2022). All sentence stimuli had the following structure: an optional carrier phrase (“Last week/Every day”) + subject and verb phrase (e.g., “She baked”) + optional number and object (e.g., “two cakes”). Across all conditions, half of the stimulus sentences include the temporal marker carrier phrase. A female speaker of American English recorded all sentences. Sound files were equalized for root-mean-square amplitude and trimmed using Audacity software (Audacity Team, 2017).

Participants used their right thumb to indicate whether the sentence was “correct” and their left thumb to indicate whether the sentence was “incorrect” as quickly as possible. Task training included an initial practice round of three to four trails with feedback from the experimenter and a practice session on seven trials on the computer. Practice sentences were all distinct from the experimental stimuli. Feedback was given during the computer practice session, and additional instructions were repeated if necessary.

Participants were seated in front of an external 23-in. Philips 230E wide LCD screen connected to a Dell OptiPlex 780 desktop computer. A cartoon alien appeared at the center of a computer screen, while the auditory stimuli were presented aurally via E-Prime software through two external speakers positioned equidistant to each side of the monitor. The task was an event-related design where the three conditions were pseudorandomized so that there were no more than four of the same condition in a row. Each trial was 5,000 ms long: The duration of sentences was approximately 3,000 ms, followed by a question mark that appeared for 2,000 ms at the center of the screen, indicating the response interval. Silent, rest periods with a fixation cross were randomly jittered for 0–6,000 ms throughout the task between trials, with jitter periods lasting approximately 25% of the total duration of the experimental task (randomized using optseq2; Dale, 1999). The task was approximately 6 min long.

The dependent variable is brain activation measured with fNIRS imaging during the grammaticality task. The fNIRS probe set contains 16 channels (light source–detector pairs) across the frontotemporal–parietal language network of the left hemisphere (see Figure 1). The regions of interest for our hypotheses are the left IFG (as measured with Channels 1–4) and the left STG/MTG (as measured with Channels 5, 7, 9, and 11; see Figure 1).

Figure 1.

Left hemisphere visualization of the functional near-infrared spectroscopy probe set configuration showing light emitters in red and detectors in blue. Black lines show all channels of data acquired. The frontotemporal areas in green (Channels 1–4, 5, 7, 9, and 11) are preregistered regions of interest and are included in the analyses. For localization details, see Hu et al. (2020).

Data Acquisition and Processing

The study used a TechEn CW6 system with 690- and 830-nm wavelengths. The fNIRS cap setup included five emitters of near-infrared light sources and eight detectors spaced approximately 2.7 cm apart, yielding 16 data “channels” per hemisphere (see Figure 1). Sources and detectors were mounted onto a custom-built head cap constructed from 2-mm silicone rubber material, with attached grommets to hold them in place during data collection. The alignment of the sources and detectors was placed precisely in a gridlike shape, yielding full coverage of the underlying regions of interest across multiple channels. The probes were applied using the international 10–10 transcranial system positioning (Jurcak et al., 2007); nasion, inion, Fpz, left and right preauricular points, and head circumference were measured; and F7, F8, T3, and T4 were anchored to a specific source or detector. TechEn CW6 software signal-to-noise ratio minimum and maximum were set to the standard 80- and 120-dB power range, respectively. Before the start of the experimental task, data quality control check was completed by looking to find the participant's cardiac signal across key channels of interest and making sure fNIRS signal in these channels fell between the minimum and maximum power parameters. If needed, trained experimenters adjusted positioning of the cap or participant's hair as necessary to detect cardiac signal. Data were collected at a sampling frequency of 50 Hz. Estimation of brain regions was calculated using the anatomical localization methods as described by Hu et al. (2020). As the preregistered report (including the hypotheses) proposed to focus on the left hemisphere, this study focuses exclusively on the left hemisphere.

Analyses

Using the NIRS AnalyzIR toolbox in MATLAB (Santosa et al., 2018), raw time course data were converted into units of optical density change. A set of customized scripts based on the study of Hu et al. (2010) were used to analyze the fNIRS data. A general linear model (GLM) framework for data analyses of the sentence processing tasks was used (Friston et al., 2006). Data analysis scripts are posted on OSF (https://osf.io/v5hkq/).

Several preprocessing steps were applied to the subject-level data, including trimming of raw data file to keep only 5 s of pre- and postexperimental task baseline data, resampling the data from 50 to 2 Hz given that the fNIRS signal of interest lies in the range of 0–1 Hz, optical density change data conversion, and hemoglobin concentration signal change data conversion using the modified Beer–Lambert law. Each participant's hemoglobin concentration data were analyzed using a fixed-effects GLM, assuming the dual-gamma canonical hemodynamic response function peaking 6 s after trial onset (Friston et al., 2006; Hu et al., 2010), yielding estimated HbO and HbR beta values for each participant, condition, and channel. At the first-level analysis, each participant's hemoglobin concentration data were analyzed using an autoregressive GLM that assumes the dual-gamma canonical hemodynamic response function (Friston et al., 2006). The first-level GLM analysis estimates beta values, which are indices of percent signal change, for each channel, sentence condition, and participant: later (ED&S), early (ING), and correct.

Group-level analyses were conducted using a linear mixed-effects model for each data channel. To correct for motion artifacts and serial correlations, we used a prewhitening autoregressive filter combined with a weighted least squares estimation approach to eliminate the nonspherical noise structure caused by physiological and motion artifacts in the time series (Caballero-Gaudes & Reynolds, 2017; Friman et al., 2004). The group-level linear mixed-effects model included task conditions (three conditions) as fixed effects, participants as a random effect variable, and hemoglobin beta values (HbO and HbR) as the predicting dependent variables. Estimated group-level beta values were extracted for each channel for each of the following contrasts: ED&S > Correct and ING > Correct. Group-level results (unstandardized beta) for each contrast were plotted on to the MNI 152 brain template using the previously specified MNI coordinates for data visualization. The HbO signal was used for all data analyses as HbO is the major contributor to the fNIRS signal (HbO, 73%–79%; HbR, 16%–22%) according to a quantification study (Gagnon et al., 2012). Additionally, several studies have found that HbR signals are particularly susceptible to noise when using the fNIRS technique (Hoshi, 2007; Strangman et al., 2002).

For Research Question 1, group analyses were conducted separately for the older and younger age groups using a second-level GLM that includes the following condition contrasts: Later > Early (errors on later-acquired morphosyntactic structures [–ed and –s] relative to an early-acquired [–ing]) and Early > Later (errors on the early-acquired morphosyntactic structure relative to later-acquired). Sentence type was a fixed effect, participants were a random effect, and hemoglobin beta value (HbO) for each contrast was the dependent variable. We controlled for task accuracy. Following the group analyses, two-tailed t tests were conducted with an alpha threshold of q < .05, controlling for multiple comparisons across number of channels measured.

For Research Question 2, group analyses were conducted using a second-level GLM that includes the following condition contrasts: Later > Early and Early > Later. Sentence type and age group were the fixed effects, participants were a random effect variable, and hemoglobin beta value (HbO) for each contrast was the dependent variables. Analyses were carried out using two-tailed t tests with an alpha threshold of q < .05, controlling for multiple comparisons across number of channels measured.

For Research Question 3, beta value (hemoglobin HbO) from the first-level analyses was extracted for all participants, regardless of age group, for the following conditions of interest: Early and Late. For each sentence type, a regression analysis was conducted in RStudio (R Core Team, 2022), with beta values as the dependent variable and the amount of English exposure (input), the amount of English use (output), and the combination of exposure and use as the independent variables.

Results

Task performance and language abilities are reported in Tables 1 and 2 for the entire group of young and older bilingual children. The young and older children did not differ in parental education, occupation, income, age of first words in English, English input, English output, combined English input/output, or English and Spanish oral language (morphosyntax and semantics) skills (p > .05).

Research Question 1

Early- and Later-Acquired Grammatical Morphemes

As a whole group, there was greater activation in the DLPFC and IFG pars opercularis for ING and the IFG pars opercularis, STG, and posterior MTG (pMTG) for ED&S (threshold p < .05; see Figure 2 and Table 3). Specifically, the DLPFC and STG were statistically significant at the corrected value q < .05, accounting for multiple comparisons.

Figure 2.

Patterns of brain activity during the grammaticality judgment task in all bilingual children for the ING > Correct (A) and ED&S > Correct (B) contrasts, controlling for age. Results are shown at p < .05 uncorrected with q < .05, corrected for multiple comparisons, marked with an asterisk.

Table 3.

Whole-group left hemisphere brain activations (unstandardized ß) for the ING > Correct and ED&S > Correct contrasts, controlling for age and overall task accuracy.

Channel	Region of interest	MNI (x y z)	ß (SE)	t stat	p uncorrected	q corrected
ING > Correct
1 IFGtri	Inferior frontal gyrus, pars triangularis	−53 37 −11	0.15 (0.21)	0.71	.476	.425
2 DLPFC	Dorsolateral prefrontal cortex	−52 38 12	0.57 (0.22)	2.58	.010	.042
3 IFGtri	Inferior frontal gyrus, pars triangularis	−57 17 −9	−0.07 (0.29)	−0.23	.822	.995
4 IFGop	Inferior frontal gyrus, pars opercularis	−57 20 14	0.49 (0.22)	2.23	.027	.071
5 STG	Superior temporal gyrus	−61 −2 −6	−0.001 (0.23)	−0.01	.995	.995
7 PAC	Primary auditory cortex	−63 −21 −6	0.01 (0.26)	0.02	.981	.995
9 pSTG	Posterior superior temporal gyrus	−62 −40 4	0.61 (0.37)	1.64	.103	.183
11 pMTG	Posterior medial temporal gyrus	−59 −58 −4	−0.31 (0.32)	−0.97	.334	.486
ED&S > Correct
1 IFGtri	Inferior frontal gyrus, pars triangularis	−53 37 −11	0.11 (0.20)	0.56	.577	.710
2 DLPFC	Dorsolateral prefrontal cortex	−52 38 12	−0.40 (0.21)	−1.87	.063	.126
3 IFGtri	Inferior frontal gyrus, pars triangularis	−57 17 −9	−0.61 (0.28)	−2.15	.033	.105
4 IFGop	Inferior frontal gyrus, pars opercularis	−57 20 14	0.16 (0.21)	0.75	.454	.606
5 STG	Superior temporal gyrus	−61 −2 −6	0.69 (0.22)	3.16	.002	.010
7 PAC	Primary auditory cortex	−63 −21 −6	−0.50 (0.25)	−1.96	.051	.116
9 pSTG	Posterior superior temporal gyrus	−62 −40 4	0.46 (0.35)	1.30	.195	.347
11 pMTG	Posterior medial temporal gyrus	−59 −58 −4	−0.77 (0.31)	−2.46	.015	.059

Research Question 2

Within-Group Comparisons for Later- Versus Early-Acquired Morphemes

Older children showed greater activation in the IFG pars opercularis (p < .05; see Figure 3 and Table 4), but this activation did not reach significance when correcting for multiple comparisons between the later- versus earlier-acquired morphemes. Younger children had significantly greater activation in the DLPFC, STG, and pMTG for ING rather than ED&S (threshold p < .05). Specifically, the DLPFC and STG regions were significant at the corrected value q < .05.

Figure 3.

Patterns of brain activity during the grammaticality judgment task in older (A) and younger (B) bilingual children for the ED&S > ING contrast, controlling for overall task accuracy. Results are shown at p < .05 uncorrected with q < .05, corrected for multiple comparisons, marked with an asterisk.

Table 4.

Left hemisphere brain activations (unstandardized ß) for the older and younger age groups for the EDS > ING contrast.

Channel	Region of interest	MNI (x y z)	ß (SE)	t stat	p uncorrected	q corrected
Old
1 IFGtri	Inferior frontal gyrus, pars triangularis	−53 37 −11	0.04 (0.28)	0.13	.899	.991
2 DLPFC	Dorsolateral prefrontal cortex	−52 38 12	−0.51 (0.30)	−1.71	.089	.472
3 IFGtri	Inferior frontal gyrus, pars triangularis	−57 17 −9	−0.94 (0.41)	−2.28	.023	.186
4 IFGop	Inferior frontal gyrus, pars opercularis	−57 20 14	−0.17 (0.30)	−0.58	.565	.991
5 STG	Superior temporal gyrus	−61 −2 −6	0.27 (0.33)	0.81	.419	.991
7 PAC	Primary auditory cortex	−63 −21 −6	−0.14 (0.41)	−0.35	.728	.991
9 pSTG	Posterior superior temporal gyrus	−62 −40 4	0.10 (0.51)	0.20	.843	.991
11 pMTG	Posterior medial temporal gyrus	−59 −58 −4	−0.06 (0.49)	−0.13	.896	.991
Young
1 IFGtri	Inferior frontal gyrus, pars triangularis	−53 37 −11	−0.04 (0.36)	−0.12	.903	.903
2 DLPFC	Dorsolateral prefrontal cortex	−52 38 12	−1.85 (0.35)	−5.23	< .001	< .001
3 IFGtri	Inferior frontal gyrus, pars triangularis	−57 17 −9	0.17 (0.44)	0.39	.701	.801
4 IFGop	Inferior frontal gyrus, pars opercularis	−57 20 14	−0.45 (0.36)	−1.27	.205	.339
5 STG	Superior temporal gyrus	−61 −2 −6	1.21 (0.34)	3.56	< .001	.004
7 PAC	Primary auditory cortex	−63 −21 −6	−0.70 (0.36)	−1.94	.054	.154
9 pSTG	Posterior superior temporal gyrus	−62 −40 4	−0.76 (0.56)	−1.35	.178	.339
11 pMTG	Posterior medial temporal gyrus	−59 −58 −4	−1.10 (0.45)	−2.46	.014	.058

Between-Groups Comparisons for Young Versus Older Children

The between-groups comparisons revealed that young children had greater activation in the DLPFC, IFG opercularis, primary auditory cortex (PAC), and posterior STG regions for ING and the IFG opercularis and PAC regions for ED&S (threshold p < .05; see Figure 4 and Table 5) relative to the older children. Specifically, the IFG opercularis and PAC regions were significant for corrected value q = .014 for ING.

Figure 4.

Patterns of brain activity during the grammaticality judgment task for the Young > Old across group contrast for the ING (A) and ED&S (B) contrast, controlling for overall task accuracy. Results are shown at p < .05 uncorrected with q < .05, corrected for multiple comparisons, marked with an asterisk.

Table 5.

Left hemisphere brain activations (unstandardized ß) for the Young > Old contrast for the ING and ED&S conditions of interest.

Channel	Region of interest	MNI (x y z)	ß (SE)	t stat	p uncorrected	q corrected
ING (Young > Old)
1 IFGtri	Inferior frontal gyrus, pars triangularis	−53 37 −11	0.29 (0.59)	0.49	.623	.808
2 DLPFC	Dorsolateral prefrontal cortex	−52 38 12	1.24 (0.60)	2.05	.042	.133
3 IFGtri	Inferior frontal gyrus, pars triangularis	−57 17 −9	−0.72 (0.74)	−0.97	.331	.538
4 IFGop	Inferior frontal gyrus, pars opercularis	−57 20 14	1.91 (0.60)	3.16	.002	.014
5 STG	Superior temporal gyrus	−61 −2 −6	−0.28 (0.64)	−0.45	.657	.808
7 PAC	Primary auditory cortex	−63 −21 −6	2.27 (0.70)	3.28	.001	.014
9 pSTG	Posterior superior temporal gyrus	−62 −40 4	2.20 (0.95)	2.30	.022	.119
11 pMTG	Posterior medial temporal gyrus	−59 −58 −4	−0.25 (0.81)	−0.31	.757	.865
ED&S (Young > Old)
1 IFGtri	Inferior frontal gyrus, pars triangularis	−53 37 −11	0.21 (0.58)	0.36	.717	.885
2 DLPFC	Dorsolateral prefrontal cortex	−52 38 12	−0.10 (0.60)	−0.17	.862	.885
3 IFGtri	Inferior frontal gyrus, pars triangularis	−57 17 −9	0.39 (0.74)	0.53	.597	.885
4 IFGop	Inferior frontal gyrus, pars opercularis	−57 20 14	1.63 (0.60)	2.72	.007	.112
5 STG	Superior temporal gyrus	−61 −2 −6	0.66 (0.64)	1.03	.303	.606
7 PAC	Primary auditory cortex	−63 −21 −6	1.71 (0.69)	2.48	.014	.112
9 pSTG	Posterior superior temporal gyrus	−62 −40 4	1.34 (0.95)	1.40	.162	.431
11 pMTG	Posterior medial temporal gyrus	−59 −58 −4	−1.29 (0.81)	−1.60	.111	.356

Research Question 3

Functional Neuroimaging Data Results: Current English Exposure and Use

Based on the group results of Research Question 1, where there was significantly greater activation of the DLPFC for ING and STG for ED&S for corrected values, these two channels were used to conduct regression analyses to identify if the amount of English exposure (input), the amount of English use (output), or the combination of exposure and use significantly predicted the beta values. For ING in the DLPFC, none of the regression models were significant (see Table 6). Therefore, exposure use nor the combination of exposure and use was able to predict activation for ING in the DLPFC.

Table 6.

Results of the regression analyses with whole-group brain activation as related to bilingual children's English language use.

Model		Estimate	SE	t	p	Adj. R 2	F	p
Left dorsolateral prefrontal cortex (Ch2 DLPFC) activation for ING > Correct
1	Intercept	0.31	5.58	0.06	.96	−.01	0.02	.88
	English input	0.01	0.08	0.15	.88
2	Intercept	−1.39	5.09	−0.27	.79	−.009	0.25	.62
	English output	0.04	0.07	0.50	.62
3	Intercept	−0.77	5.51	−0.14	.89	−.01	0.12	.73
	English input and output	0.03	0.08	0.35	.73
Left superior temporal gyrus (Ch5 STG) activation for ED&S > Correct
1	Intercept	12.82	5.69	2.25	.03	.04	4.70	.03
	English input	−0.18	0.08	−2.17	.03
2	Intercept	8.95	5.26	1.70	.09	.02	2.58	.11
	English output	−0.12	0.07	−1.61	.11
3	Intercept	11.44	5.65	2.03	.05	.03	3.76	.06
	English input and output	−0.15	0.08	−1.94	.06

For ED&S in the STG, only the regression model with English exposure predicted the beta values, F(1, 79) = 4.70, p = .03, and explained 4.42% of the variance (see Table 6). Therefore, exposure was the only variable that was able to predict activation for ED&S in the STG. However, none of the models survive a Bonferroni correction for multiple comparisons (p = .02). The regression model with English exposure is trending toward significance and shows that as English input increases, beta values slightly decrease.

Discussion

This study investigated the neurocognitive bases of early-acquired (–ing) and later-acquired (–ed and –s) English grammatical morphemes in young Spanish–English bilinguals and how functional activation varies with age and current English use. In children ages 6–11 years, we observed activation in the left DLPFC, IFG, MTG, and STG during an auditory morphosyntactic grammaticality judgment task, in line with prior work on monolingual children. In younger children (6;0–8;11), there was activation in the left DLPFC, STG, and pMTG. In older children (9;0–11;11), there was activation in the left IFG. Patterns of brain activation for morphosyntactic processing in the left STG were related to bilingual children's English language exposure.

First, we examined the neural correlates of early- and later-acquired English morphosyntactic structures in Spanish–English bilingual children as a group. We predicted that the left IFG and left STG/MTG would be primarily recruited for morphosyntactic processing during sentence comprehension with the strongest activation in the IFG pars opercularis and pars triangularis. This prediction was based on theoretical and empirical work for language and brain development positing that the left IFG and STG are part of the dorsal stream language neurocircuitry, which, over time, becomes fine-tuned for morphosyntactic processing (e.g., Enge et al., 2020; Skeide & Friederici, 2016; Wagley et al., 2023; Wang et al., 2020, 2021; Wang, Wagley, et al., 2022). The findings are generally consistent with our predictions and the guiding literature. For the early-acquired grammatical morpheme, present progressive (–ing), the left DLPFC and left IFG pars opercularis were primarily recruited. For the later-acquired grammatical morphemes, past tense and third-person singular (–ed and –s), the left STG was primarily recruited. These findings are generally consistent with prior works on monolingual children of the same age that report similar activation patterns in the left frontal and temporal lobes during syntactic processing (Friederici, 2009; Friederici et al., 2017; Skeide, 2012; Wang et al., 2021).

Our results are slightly inconsistent with prior neuroimaging studies on Spanish–English bilingual children. Arredondo et al. (2019) found that both monolingual and bilingual children showed greater neural activity in the left IFG for later-acquired structures (tense/agreement) than earlier-acquired structures (–ing). While in another study, Wagley et al. (2023) found significant activation in the left IFG (including pars opercularis, triangularis, and orbitalis regions) for later-acquired structures (–ed and –s). In this study, the frontal lobe was recruited for the early-acquired structures (–ing) but not the later-acquired structures (–ed and –s). Although all of these studies focused on Spanish–English bilingual children around the same ages (ranging from 6 to 12 years) due to the heterogeneity of bilingualism and childhood experiences with language, there are differences in neural activity in response to morphosyntactic processing. Thus, it is vital to continue to operationalize definitions and clearly explain measures being used so that we can compare across studies in the future as the pediatric bilingual neuroimaging literature base continues to grow (Surrain & Luk, 2019).

Second, we investigated if the neural correlates of early- and later-acquired English morphosyntactic structures varied with age. We predicted that older children would have more restricted activation in the left IFG for earlier-acquired (–ing) and later-acquired (–ed and –s) morphosyntactic structures in comparison to more distributed activation in younger children for the later-acquired structures (–ed and –s). Older children did indeed have activation focused in the left IFG pars triangularis, showing more specialized and focal IFG engagement for syntactic processing (e.g., Brauer & Friederici, 2007; Nuñez et al., 2011; Román et al., 2015; Skeide et al., 2014; Sulpizio et al., 2020; Wang et al., 2020, 2021; Wu et al., 2016). Younger children had more distributed activation including the left DLPFC, STG, and pMTG for the later (–ed and –s) compared to earlier (–ing) structures. When comparing directly between the two age groups during the earlier-acquired (–ing) sentence processing, the younger children had greater activation in four regions of interest, the left DLPFC, IFG pars opercularis, PAC, and posterior STG. Similarly, during –ed and –s omission processing across groups, younger children had greater activation in the left IFG pars opercularis and PAC. This widespread activation in younger children has been noted in previous work focusing on syntactic processing (e.g., Knoll et al., 2012; Luke et al., 2002; Skeide et al., 2014, 2016; Skeide & Friederici, 2016) and likely demonstrates more neural recruitment. In summary, in support of theoretical frameworks positing that brain specialization for language develops in temporal regions first, older children had more restricted activation than the younger children across both early-acquired (–ing) and later-acquired (–ed and –s) morphemes, demonstrating a more adultlike and perhaps more automated cortical response in the frontal region in which linguistic representations are already encoded and more efficiently accessed (e.g., Arredondo et al., 2019; Enge et al., 2020; Skeide et al., 2014, 2016).

Last, we examined if the neural correlates of early- and later-acquired English morphosyntactic structures varied as a function of current English exposure and use. We predicted that bilingual children with more current English exposure and use would have more restricted neural activity in both the left IFG and left STG/MTG for earlier-acquired (–ing) and later-acquired (–ed and –s) grammatical structures than children who have less English exposure and use. We found that current English exposure and/or use was not related to the activation for the early-acquired structure –ing. However, for the later-acquired structures –ed and –s, the relation between English exposure was trending toward significance in the left STG, suggesting that bilingual children who have more English exposure may be relying on less neural engagement during English sentence processing. As bilingual children hear a language more often, with the typical morphosyntactic constructions, their language processing becomes more efficient and less effortful. Thus, they may need less neural recruitment to process these grammatical structures due to the amount of English exposure each day. This is counter to some of the behavioral bilingual research that has found that both exposure and use predict behavioral morphosyntactic performance in each language (Bohman et al., 2010). The morphosyntactic behavioral measures, within the Bohman et al. (2010) study, were all productive in nature where a verbal response was required for each item. It is possible that since the fNIRS experimental grammaticality judgment task was receptive and no overt verbal response was necessary, the task more closely aligns with language exposure as it focuses on the amount of language heard on a regular basis.

Limitations and Future Directions

One limitation of this study is the way the stimuli are presented. The sentences were created with omissions as the manner in which to test morphosyntactic processing (e.g., Tokowicz & MacWhinney, 2005). In the natural environment, people typically do not hear these many sentences with omission errors. This may lead to an additional cognitive load for error processing (Wagley et al., 2019). Sentence tasks build on an established body of literature; however, these types of tasks rely on isolated sentences with reduced prosodic features and are out of context. In the future, researchers are encouraged to use story listening tasks as they have higher ecological validity (Brennan, 2016).

Second, these findings focus on grammatical morphemes in English in bilingual children. Bilinguals learn each of their languages in different contexts, and children vary in the amount and types of experiences they have in each language. Behaviorally, bilingual speech-language pathologists assess both languages to obtain a clearer picture of the child's skills across languages and across domains as there are distinct linguistic markers of impairment that differentiate bilingual children with and without DLD in each of their languages. Thus, in the future, it would be beneficial to investigate morphosyntactic processing across both languages using neuroimaging to better understand patterns of activation.

Last, the focus and hypotheses of this study centered around the left hemisphere. Adults tend to exhibit more left lateralization for language; however, future studies that focus on bilingual children should include both hemispheres in their analyses. There have been several studies that have found bilateral activation during expressive and receptive language tasks (e.g., Cargnelutti et al., 2019; Enge et al., 2020; Sulpizio et al., 2020).

Conclusions

This study aimed to examine the influence of bilingual language exposure on the brain organization for morphosyntactic structures. In summary, the findings for young bilingual learners are generally commensurate to those previously suggested for monolinguals: Children's neural response was strongest to later- than earlier-acquired elements of syntax. This response was more focal within the left hemisphere in older speakers and related to the amount of day-to-day English exposure. Association between English experience and left STG functionality for the processing of the more analytically complex and phonologically subtle morphemes further highlights the importance of language usage in relation to children's sensitivity and experience with the central elements of morphosyntax.

As more and more people worldwide continue to speak two or more languages, it is vital to better understand the impact bilingualism has on language development in the brain. Our findings add to a steadily growing literature on bilingualism and its effects on brain development and patterns of activation. As past tense and third-person singular constructions are red flags for DLD, the better we can characterize the neurotypical bilingual brain, the better we can understand the differences we may encounter in children with DLD and other developmental disabilities.

Author Contributions

Alisa Baron: Conceptualization (Lead), Formal analysis (Equal), Writing – original draft (Lead), Writing – review & editing (Equal). Neelima Wagley: Data curation (Lead), Formal analysis (Equal), Investigation (Lead), Software (Supporting), Supervision (Lead), Writing – review & editing (Equal). Xiaosu Hu: Data curation (Supporting), Methodology (Supporting), Software (Lead). Ioulia Kovelman: Funding acquisition (Lead), Methodology (Lead), Supervision (Supporting), Writing – review & editing (Equal).

Data Availability Statement

The data set analyzed for this study is available from the corresponding author on reasonable request.

Funding Statement

This work was supported by National Institute of Child Health and Human Development Grants R01HD092498, awarded to Ioulia Kovelman, and a department training grant T32HD007109, awarded to Neelima Wagley. This collaboration was supported in part by the American Speech-Language-Hearing Association's Advancing Academic-Research Careers Award, awarded to Alisa Baron.