Content area
Understanding how children acquire syntactic structures from a limited set of grammatical rules and use them creatively to convey meaning has been a longstanding interest for scientific communities. Previous studies on syntactic development have revealed its close correlation with the development of vocabulary and working memory. Our study sought to elucidate how the relations between syntactic processing, word processing, and working memory were instantiated in the brain, and how earlier neural patterns might predict language abilities one year later. We employed functional near-infrared spectroscopy to examine among preschool children (N=50, M
Human language represents a distinct cognitive capability that enables effective communication among individuals. Within the intricate structure of language, syntax emerges as the fundamental constituent facilitating the combination of words into coherent phrases and sentences ( Hauser et al., 2002; Berwick et al., 2013). Between the ages of about 3 and 6, children undergo a pronounced acceleration in developing their syntactic competencies, characterized by a progression from simple to more complex syntactic structures ( Clark, 2009). Understanding how children acquire syntactic structures from a limited set of grammatical rules and use them creatively to convey meaning has been a longstanding interest for many scientific communities, including psychologists, educators, linguists, and so on.
Research on syntactic development suggests children acquire simpler syntactic structures before more complex ones. They typically begin with basic sentence structures. However, this does not necessarily conform to a fixed word order, as syntactic acquisition can vary depending on the language's typological characteristics. Some languages may naturally use orders such as subject-verb-object (SVO), subject-object-verb (SOV), or verb-subject-object (VSO). Regardless of the specific order, simpler structures are generally mastered earlier than more complex syntactic forms such as relative clauses and passives. This pattern of acquisition has been noted across various languages, suggesting the possibility of universal principles in syntax learning ( Nowak et al., 2001; Yang et al., 2017). Moreover, the pace and pattern of acquiring syntactic structures may be influenced by factors including children's vocabulary size ( Bates & Goodman, 1997), cognitive abilities such as working memory ( Gooch et al., 2016; Romeo et al., 2022), and social-environmental factors ( Huttenlocher et al., 2010).
The critical mass hypothesis proposed by Bates and colleagues highlights the importance of vocabulary in the acquisition and development of syntax ( Marchman & Bates, 1994; Bates & Goodman, 1997). This hypothesis posits that children must acquire a sufficient amount of vocabulary to comprehend and utilize syntactic rules, implying that children's syntactic development depends on the growth of the lexicon. This idea has received support from studies suggesting a strong correlation between vocabulary and grammar development ( Marchman & Bates, 1994; Bates & Goodman, 1997; Devescovi et al., 2005; Dixon et al., 2007). For example, cross-sectional studies have revealed a positive association between children's vocabulary and grammar in different languages between the ages of 2 and 5 ( Devescovi et al., 2005; Pérez-Leroux et al., 2012). Meanwhile, longitudinal studies have found that the size of children's early vocabulary predicts their subsequent syntactic development ( Bates et al., 1988, Moyle et al.,2007). On the other hand, some other research suggests that an understanding of syntax can also drive the expansion of vocabulary. The syntactic bootstrapping hypothesis, for example, posits that children can use syntactic cues to infer the meanings of new words ( Landau & Gleitman, 1985; Naigles, 1990). Additionally, recent studies have cast doubt on the strength of the direct relationship between vocabulary size and syntactic development, suggesting that while a relationship exists, it may not be as deterministic as once thought, and other factors may be also at play in the complex process of language development ( Hoff et al., 2018; Brinchmann et al., 2019).
Moreover, cognitive factors such as working memory have also been identified as influential contributors to children's syntactic development. Although the debate remains unresolved over whether there are specialized working memory resources for syntactic processing, separate from those employed in general verbal working memory ( Caplan & Waters, 2013; Fedorenko et al., 2007; Just & Carpenter, 1992; Matchin, 2018; Novick et al., 2005; Santi & Grodzinsky, 2007), there is compelling evidence that sentence processing, particularly of complex syntactic constructions like relative clauses, relies on the capacity of working memory to hold and manipulate linguistic information for integration and comprehension ( Gibson, 1998; Caplan & Waters, 2013). From a developmental perspective, children's development of syntactic ability is closely tied to their enhancement of working memory, and these two key components of development typically progress hand in hand as children age. For example, the GEM (Gillam–Evans–Montgomery) model proposed by Montgomery and colleagues ( Montgomery et al., 2018, 2021) has highlighted the key role of working memory in children's development of syntactic sentence comprehension. Previous studies have demonstrated that working memory is strongly correlated with the comprehension of complex sentences in preschoolers ( Boyle et al., 2013; Verhagen et al., 2016) and school-age children ( Montgomery et al.,2008). In addition, children with better verbal working memory capacities demonstrate superior abilities in processing complex sentences ( Delage et al., 2019). Thus, working memory may contribute to children's ability to comprehend sentences, particularly on structures that have higher memory demands.
Recent years have witnessed an increase in studies investigating the neural basis of syntactic development. These studies have generally indicated that brain activation in the frontal and temporal regions is correlated with syntactic development in children ( A. Baron et al., 2023; Enge et al., 2020; Skeide et al., 2014; Skeide et al, 2016; Wagley et al., 2024; Wu et al.,2016). These investigations reveal that the frontotemporal network shows heightened activation during the processing of increasingly complex syntactic structures. Furthermore, this activation undergoes developmental changes, with more pronounced activity or more complex interactions observed in adults compared to children. For example, Wu et al., (2016) found that 5-year-old native German speaking children demonstrated greater activation in the left middle frontal gyrus (MFG) and caudate nuclei when processing complex sentences (non-canonical object-first sentences) as compared to simpler sentences (canonical subject-first sentences). In another fMRI study, Skeide et al., (2014) manipulated syntactic complexity and semantic plausibility using subject-relative and object-relative clauses. They found that children in the 3–4 age group did not exhibit significant main effects of syntax or semantics in any brain region. In contrast, children between 6 and 7 began showing semantic and syntactic effects in the left superior temporal gyrus (STG), indicating a more established neural mechanism for language comprehension at this later stage.
Although previous studies have shed light on the intricate relations between vocabulary, working memory, and syntax acquisition, there remains a dearth of research on how the connections are instantiated in the brain. Few studies have addressed this question. For instance, Fengler et al. (2016) investigated how brain maturation and verbal working memory capacity related to the processing of complex sentences in different age groups of subjects (5–6 years, 7–8 years, and adults). They found that the activation of the syntactic process in the left inferior parietal lobe and the posterior STG was predicted by the common contribution of structural brain maturation and verbal working memory capacity. These findings suggest that children's syntactic ability is intricately tied to brain maturation ( Petersen, 2017) and working memory capacity. Yet this study did not measure neural activity associated with verbal working memory directly, and thus, could not establish a brain-level connection between working memory and syntactic processing, which is vital for understanding how working memory promotes syntactic development at the brain level.
In this study, we sought to unravel the neural correlates and intricate relationships between syntactic development, word comprehension, and working memory, as well as how earlier neural patterns might serve as indicators for future development in language abilities one year later. The theoretical motivation for linking these factors derives from models of language acquisition that propose language comprehension is not merely a matter of knowing words but also involves the dynamic manipulation and integration of linguistic information. Working memory is crucial for this integration, as it provides the cognitive space in which words can be held, processed, and structured into meaningful sentences according to syntactic rules ( Gathercole, & Baddeley, 2014; Just & Carpenter, 1992). We hypothesized that developments in lexical-semantic and working memory capacities played pivotal roles in children's syntactic development. We expected that this relationship may manifest as identifiable correlations within specific neural regions associated with each cognitive function. Drawing on existing research, which highlights the importance of the posterior temporal cortex in lexical-semantic processing and the roles of the inferior and middle frontal cortex in working memory, we proposed that these regions may bridge connections of syntactic processing with lexical-semantic processing and working memory during sentence comprehension. Furthermore, we hypothesized that activity level and the effective integration of the key regions within the neural network were crucial for enhanced language comprehension. We expected that brain activity and connectivity patterns among these regions would serve as vital indicators for predicting future language development.
We employed functional near-infrared spectroscopy (fNIRS) to capture the neural activation associated with processing sentences of varying syntactic complexities, as well as tasks assessing word comprehension and working memory. fNIRS is a non-invasive and child-friendly technology widely used to study neural activity and functional connectivity in children ( Wilcox et al., 2015), with good reliability and validity ( Pinti et al., 2020). In addition, we administered a battery of behavioral tests to evaluate each participant's sentence comprehension ability, vocabulary size, and working memory. These assessments were implemented twice over the course of a year. This longitudinal approach allowed us to establish a predictive model based on the neural data collected in the first year to predict children's language performance in the following year.
We had three principal aims. The first aim was to examine whether there were any differences in how children's brains reacted and interacted when processing sentences of varying complexities, and how their brain activity related to their behavior performance. We expected to see an effect of syntactic complexity in brain activation and functional connectivity of frontotemporal regions. The second aim focused on the brain instantiation of the ties between syntactic development and factors of word comprehension and working memory. We correlated brain activation of sentence processing with brain activation evoked by word processing and working memory. The rationale for this analysis is that if a strong correlation is identified, it would suggest that these cognitive functions may share an underlying neural basis and are closely interlinked ( Thibault et al., 2021). The third aim was to leverage the neural data from the initial year to construct predictive models of children's future language achievements based on activity and connectivity patterns. We used machine learning methods to achieve this, as these algorithms can build predictive models by analyzing multivariate features of brain function, thereby maximizing individual variability in brain imaging data. By incorporating built-in cross-validation strategies and using independent data for fitting and prediction, this approach prevents overfitting and ensures effective generalization of the results, highlighting the association between brain activity and behavior ( Rosenberg et al., 2018).
2 Method2.1 Participants
A total of 67 children completed both behavioral testing and the fNIRS experiment at Time 1 (T1). They were recruited from K2 and K3 (the second and third year of kindergarten level) in a kindergarten in Shenzhen, China. All children had normal hearing, normal vision, or corrected vision, and no known language disorders. Our sample size was informed by the G*power calculation with 90% power and 25% effect size, which yielded a recommendation of 36 participants. To bolster the statistical power and ensure a robust cohort for our longitudinal study, we expanded the initial sample size beyond this recommendation. All the participants were native speakers of Mandarin, the official dialect of mainland China, and the language of instruction in kindergarten. Among these children, 17 participants were excluded due to failure to complete the tasks or too much head movement during fNIRS experiment. The remaining 50 children were included in the final analyses at T1 (33 boys, Mean age = 61.5 months, SD = 6.10 months, range=51–75 months). One year later (Time 2, T2), behavioral data were collected from 34 of the 50 children (19 boys, M = 70.2 months, SD = 2.79 months, range = 63–74 months). The relatively high dropout rate in our longitudinal study occurred due to the educational transition of some children from kindergarten to primary school, and the constraints imposed by the COVID-19 pandemic, which made data collection at the second testing phase challenging. The study was approved by the Ethics Committee of Shenzhen University. Parents of each participating child signed a written informed consent before the experiment.
2.2 Behavioral assessmentBehavioral tests were administered individually at T1 and T2. An identical set of tests was used to maintain uniformity and comparability across both sessions. These assessments were conducted outside the scanner.
2.3 Sentence comprehension testTo evaluate children's sentence comprehension abilities, we employed a sentence-picture matching task designed as a reliable index of comprehension with a focus on syntactic aspects. This test was based on the "Language proficiency test for preschool children" developed by the Language Research Institute of Tianjin Normal University (2016). This assessment incorporates a diverse range of syntactic structures to ensure a comprehensive evaluation, including simple sentences, various types of clauses, negations, and passive constructions. During the task, an experimenter read a sentence aloud to the children and presented them with two pictures. The children were then asked to select the picture that most accurately represented the meaning of the sentence they heard. A practice item was administered to ensure that the children understood the task before proceeding to the formal assessment. The formal test contained 35 items, and the children's scores were calculated based on the number of correct answers.
2.4 Vocabulary testsWe used a picture identification task and a picture naming task to measure children's receptive vocabulary and expressive vocabulary, respectively. The two vocabulary tests were the subtests of the Chinese version of the Wechsler Preschool and Primary Scale of Intelligence, Fourth Edition (WPPSI-IV) ( Li & Zhu, 2014). In the picture identification task, participants were presented with four pictures and asked to select the one that matched the word spoken by the experimenter. This test consisted of 31 items. In the picture naming task, participants were asked to name the object that corresponded with the picture presented by the experimenter. This test included 24 items. The test scores were the number of items correctly answered.
2.5 Working memory testsForward-digit span and backward-digit span tests were used to measure children's working memory. The forward-digit span test mainly measures phonological loop capacity, and the backward-digit span test assesses both central executive and storage of working memory ( Gathercole et al., 2004). The digit span test used in this study was custom-designed. In the forward digit span task, participants were asked to repeat a series of digits ranging from 2 to 9 in length. In the backward digit span task, they were asked to repeat the digits in reverse order. Each sequence length had three items. The sequence length was increased if children recalled one item correctly at that length, and only if they failed to recall all three items, the test was terminated. In the scoring of the sequence series, each digit correctly recalled in its precise position was scored one point. For example, if a child repeated the sequence "1, 2, 2, 4″ in response to the provided sequence "1, 2, 3, 4″ during the forward digit span task, he or she would be credited with three points, as the first, second, and fourth digits were correctly replicated. This scoring method is a continuous measure that can better capture individual differences among children, and has higher reliability compared to the scoring of the maximum length of digit span ( Friedman et al., 2005).
2.6 Socioeconomic status (SES) measurementThe SES index can reflect children's family environment and background. Previous studies have demonstrated that material components of the SES, such as income and education, showed the strongest association with children's language and cognitive development ( Webb et al., 2017). In this study, we measured children's SES using a questionnaire completed by their parents, and it assessed parents’ education levels (reported based on 6 levels: 1 = primary school or no formal education; 2 = middle school; 3 = high school; 4 = college or associate's degree; 5 = bachelor's degree; 6 = master's or doctor's degree) and the monthly family income (reported based on 6 levels: 1 = below ¥10,000; 2 = ¥10,000∼20,000; 3 = ¥20,000∼30,000; 4 = ¥30,000∼40,000; 5 = ¥40,000∼50,000; 6 = above ¥50,000). The parents in the sample had an average educational attainment equivalent to a bachelor's degree. The mean monthly income of these families fell within the range of ¥ 30,000 to 40,000, which is slightly higher than the median monthly income for the broader population in the city ( Shenzhen Municipal Bureau of Statistics, 2022). We performed a z-score transformation on the education and income data and then summed the transformed average education levels of both parents with the family income to get the composite SES score.
2.7 fNIRS tasks and procedureParticipants completed the fNIRS tasks at T1.
2.8 Picture-matching task for sentence and word comprehensionWe employed picture-matching tasks to examine brain activation patterns associated with sentence and word comprehension. This paradigm is commonly employed to investigate children's language development in behavioral and neuroimaging studies ( Skeide et al., 2014; Kidd and Arciuli, 2016). By having participants engage in tasks at sentence and word levels, we could evaluate the brain's response to both syntactic processing inherent in the sentence-level task, and lexical-semantic processing, which is a key aspect of the word-level task. In the picture-matching task, participants were presented with two pictures, one on the left and the other on the right side of the screen, accompanied by audio stimuli that delivered either a sentence or a word. The presentation of the correct picture had an equal probability on the left and right sides of the screen. The children were instructed to select the picture that matched the audio by pressing the keyboard ( Fig. 1a ). For example, when hearing the sentence “小牛踢了小羊” (meaning "The cow kicks the lamb"), participants were shown one picture of a cow kicking a lamb and another picture of a lamb kicking a cow.
In the picture-matching task for sentence processing, the choice of animals and verbs within the sentences was made considering the frequency of their use among children, according to the corpus study of Chinese children ( Zhou et al., 2021). In total, we generated 48 unique sentences using a combination of eight animal nouns and eight distinct verbs. There were three sentence conditions that varied in syntactic complexity, i.e., canonical SVO sentence “小牛踢了小羊” (“The cow kicks the lamb”), passive sentence “小羊被小牛踢了” (“The lamb was kicked by the cow”), and subject relative clause (SRC) “踢了小羊的小牛” (“The cow that kicks the lamb”). In the SRC condition, we presented stimuli that were not complete sentences, but they still conveyed clear and grammatically acceptable meanings. This design ensured the consistency of stimulus length and information content across different sentence conditions (e.g., Boyle et al., 2013). The selection of passive sentence and SRC was mainly based on two considerations: the specific challenges these structures pose in children's syntactic development and the need to control sentence length across different conditions. Passive and SRC, with their non-canonical word order and elevated cognitive demands, represent a substantial developmental leap in language acquisition. These structures often emerge later in language development due to their complexity. In our study, choosing syntactic forms that could be processed with reasonable accuracy and speed was essential. We aimed to use sentence structures that were sufficiently challenging yet comprehensible within the children's developmental stage. Additionally, passive and SRC allow for increased syntactic complexity while maintaining sentence length, as their constructions can incorporate additional syntactic elements without excessively extending the sentence. Each syntactic condition consisted of 16 trials, resulting in a total of 48 sentences. A block design was used, with each condition containing four blocks presented in a pseudo-random order. There were four trials in each block that lasted for 30 seconds. Each trial started with a fixation randomly jittered at 500ms, 1000ms, or 1500ms to prevent habituation and maintain participants' engagement. After the fixation, a picture and an auditory sentence were presented to the participants for 6500ms, during which they made their responses.
In the word comprehension task, the children were instructed to select the picture that matched the word by pressing the keyboard. The word stimuli in this task encompassed eleven diverse categories of common objects familiar to children, such as fruits, vegetables, and vehicles. There were four blocks, and each block lasted for 30 seconds and contained six trials. Each trial began with a fixation randomly jittered at 500ms, 1000ms, or 1500ms and the duration of the picture and auditory word presentation was 4000ms. This task was similar to that used in the receptive vocabulary test, but instead of assessing the size of the children's vocabulary, it primarily examined the neural responses elicited during word comprehension. We used the term "word comprehension" in the fNIRS section to differentiate it from the outside-scanner vocabulary tests.
The formal picture-matching task took around 14 minutes to complete, with a rest break of 1–2 minutes provided to the children after they finished half of the task. Prior to the formal task, a practice session was conducted to ensure comprehension, with the formal experiment commencing only after the participants achieved a correct response rate exceeding 80%. The audios of stimuli were recorded by a native Chinese female speaker in a child-directed way and were digitized and normalized to the root mean squared amplitude.
2.9 The N-back task for assessing working memoryWe used an n-back task to assess children's verbal working memory during fNIRS recording. The task involves presenting participants with a sequence of auditory stimuli, and the participants are required to determine if the current stimulus matches the one from 'n' steps earlier in the sequence. In our study, the task included a 1-back condition and a 0-back condition ( Fig. 1b). In the 1-back condition, the participants were instructed to press the keyboard whenever the current number matched the number that immediately preceded it. This condition required maintaining and updating information in memory. In the 0-back condition, the participants were instructed to remember a target number (the number 5) and pressed the keyboard when they heard the target number. To avoid practice effects, the target number in the practice section was different from the target number in the formal experiment. The 0-back condition serves as a control condition since it doesn't require updating information in memory. It enabled us to control cognitive components such as basic attention, processing speed, and response inhibition. The children performed the 0-back task, followed by the 1-back task. Each task started with a 30-second fixation period. This was followed by four task blocks; each block initiated with a 1-second preparation period and comprised 12 trials, each lasting 2500 ms, totaling 31 seconds per block. Upon completing the 0-back task, children were given a rest period, which typically lasted about two minutes. The total duration of the entire n-back task was approximately 8 minutes. There was a practice section before the formal experiment, and the formal experiment began when the participants achieved a correct response rate above 80%.
Most of the children completed the picture-matching task and n-back task in the same session, and two children took part in the two tasks in two separate sessions a few days apart because they felt it difficult to concentrate for the length of time required to complete both tasks in one session. In addition, ten children were unable to complete the n-back task. Therefore, the analysis of working memory brain activation included only the remaining 40 participants. All experimental tasks were presented via E-prime 3.0 on a DELL monitor with a view distance of about 60 cm from the monitor.
2.10 Imaging acquisitionThe fNIRS signal acquisition was recorded in a child-friendly room. Children were comfortably seated in a chair and instructed to stay silent and to avoid moving as much as possible. The hemodynamic responses were measured with a multiple-channel fNIRS system (Oxymon MK Ⅲ, Artinis, The Netherlands), with two wavelengths (760 nm and 850 nm) measuring the changes in optical density, which was converted to the changes of the concentration of oxygenated hemoglobin (HbO), deoxygenated hemoglobin (HbR) and total hemoglobin (HbT). The sampling rate was set at 10Hz.
There were 16 detectors and 16 receivers, forming 48 measurement channels (24 per hemisphere) that broadly covered parts of the bilateral frontal and temporal cortices. Each paired detector and receiver were separated by a 2.5 cm distance. The array positions were determined based on the 10–20 system ( Jasper, 1958). We specifically referred to F7, F8, T3, and T4 to cover the frontal and temporal areas. F7 and F8 correspond to the left and right inferior frontal areas, while T3 and T4 correspond to locations over the left and right temporal regions. This arrangement ensures optimal coverage of key areas involved in language comprehension. A three-dimensional digitizer (Polhemus Corp.) was used to measure the coordinates of all the optodes, along with five additional anatomical references (nasion, inion, left and right preauricular points, and Cz) for some children. These channel positions were then transformed into Montreal Neurological Institute (MNI) space using an affine transformation matrix derived from reference MRI data ( Singh et al., 2005). These coordinates were superimposed on a cerebral cortex atlas using the statistical parametric mapping for near-infrared spectroscopy (NIRS-SPM) toolbox ( Ye et al., 2009). When fitting the caps on the children, the Cz optode was positioned at the midpoint between the nasion and inion, aligning with the preauricular points.
3 Data analysis3.1 Behavioral data analysis
We first calculated the scores of the behavioral assessments (i.e., sentence comprehension, vocabulary, and digit span tests), and the mean accuracy rate (ACC) of the fNIRS tasks (i.e., sentence processing, word processing, and n-back tasks). We conducted a repeated-measure ANOVA on the behavioral performance of the three syntactic sentence conditions in fNIRS task and a paired-sample t-test on the 1-back and 0-back conditions of the working memory task. We then applied Pearson correlation analyses to examine the relations among these behavioral measurements. For the longitudinal data, we conducted paired-sample t-tests to compare the scores of sentence comprehension, vocabulary, and working memory at T1 and T2, to assess the developmental change in language and cognitive abilities.
3.2 fNIRS data pre-processing and activation analysisFNIRS data were analyzed using the NIRS AnalyzIR toolkit ( Santosa et al., 2018). First, the raw data was downsampled from 10 Hz to 4 Hz, and then the optical density signal was converted to a blood oxygen concentration signal based on the Beer-Lambet law. We used an autoregressive iteratively reweighted least-squares (AR-IRLS) method ( Barker et al., 2013) to remove head movement artifacts and physiological noise of respiration and heartbeat ( Yücel et al., 2016). This method is effective in controlling systemic physiology and motion-related artifacts ( Huppert, 2016). To estimate the activation strength of each condition, we convolved canonical hemodynamic response functions with a boxcar model. This allowed us to compute the estimated regression coefficients (beta values) of each regressor in the general linear model, which represented the activation strength of each channel for each condition. Group-level analyses were performed using a linear mixed-effects model. To examine the activation across conditions, we fitted models with the condition as a fixed effect and subject as a random effect, the corresponding formula was "beta∼ condition+(1|Subject)". All statistics were based on the HbO signal, which has a higher signal-to-noise ratio than the HbR signal and can better avoid the interference of noises ( Hoshi, 2003). The results were corrected by multiple comparisons using false-discovery rate (FDR) correction with q < 0.05 ( Benjamini and Hochberg, 1995).
3.3 Brain-behavior correlation analysisNext, we applied linear regression to examine whether brain activation during sentence processing was correlated with children's language performance. In this analysis, we focused on the brain regions significantly activated during sentence processing in the picture-matching task. We defined an ROI as a brain area containing at least two channels that exhibited significant activation during sentence processing. This criterion was established to ensure robust signal detection and to minimize the noise typically associated with single-channel data. As shown in Fig. 2 , these included the left inferior frontal gyrus (left IFG, channels marked by green circle), left middle and superior temporal gyrus (left MTG/STG, channels marked by blue circle), and right superior temporal gyrus (right STG, channels marked by red circle). Each of these ROIs consistently contained two activated channels, aligning with our criteria for reliable signal detection. The mean beta values extracted for each ROI across the three sentential conditions were transformed to z-score and used as independent variables. We used the fNIRS-task performance for each sentential condition, and the sentence comprehension scores from cognitive-behavioral measures as dependent variables. To control for the effects of age and gender, they were included as covariates in the analysis.
3.4 Brain functional connectivity (FC) analysisLanguage comprehension is a complex cognitive process supported by distributed brain regions ( Fedorenko et al., 2014). Previous research has shown that the resting-state functional connection between the left IFG and the left posterior superior temporal sulcus (pSTS) in young children is related to their ability to comprehend more complex sentences ( Xiao et al., 2016). In this study, we examined whether FC between frontal and temporal regions differed among the three syntactic conditions during sentence processing and whether the FC strength was correlated with children's sentence comprehension performance. We chose the left IFG, the left STG/MTG, and their right-side homologous regions as ROIs. It should be noted that while the right IFG did not show significant activation in our activation analysis, this does not preclude its involvement in language development through connectivity pathways. It could play a role in overall network dynamics through their connections with other brain regions. Therefore, including the right IFG in our connectivity analysis allows us to examine both intra- and interhemispheric connectivity to explore the possibility of such contributions.
We employed an iterative autoregressive least-square method to calculate the estimated correlation coefficients for the FC between ROIs ( Santosa et al., 2017). In this analysis, an ROI is defined as a group of channels within a specific brain region. These included the left IFG (channels 2, 3, 7, 8, 15, 16, 21); right IFG (channels 26, 27, 31, 32, 39, 40, 45); left STG/MTG (channels 14, 17, 18, 22, 23, 24); right STG/MTG (channels 38, 41, 42, 44, 46, 47, 48). We calculate the correlation coefficients among channels of these ROIs to obtain the FC values. The FC value between two ROIs was derived from the correlation coefficients of channels in one ROI with those in the other ROI. In addition, we conducted a linear regression analysis to examine the relation between brain functional connectivity and language performance, with connectivity strength as the independent variable, and language performance as the dependent variable. Age and sex were included as covariates in the analysis.
3.5 Relation between sentence processing, word processing, and working memory in the brainTo examine how the relations among syntactic processing, word processing, and working memory were instantiated in the brain, we analyzed the correlations between brain activity during sentence processing with brain activity evoked by word processing and working memory. This approach has been used in previous neuroimaging studies ( Lacey et al., 2014; Thibault et al., 2021). We used the brain regions that were found critically involved in sentence processing as ROIs and analyzed the connections between syntax, vocabulary, and working memory based on these ROIs. These ROIs included the left IFG, left STG/MTG, and right STG, as illustrated in Fig. 2. We extracted beta values from these ROIs for the sentence condition, word condition, as well as the 1-back and 0-back conditions in the working memory task, and then standardized them using a z-score transformation. Using Pearson correlation, we calculated the correlations between brain activity in the sentence and word conditions, as well as the working memory (1-back>0-back) condition, for each ROI. In this analysis, we excluded data that fell outside of three standard deviations from the mean.
3.6 Prediction of T2 language performance based on neural data at T1Our analyses of data at T1 revealed the effects of syntactic complexity on both brain activation and functional connectivity. We therefore developed models based on brain activity and connectivity patterns within the left IFG, STG, and their right homologous regions to predict children's future language abilities. For the longitudinal experiment, we implemented Support Vector Regression (SVR) ( Scholkopf, 1999) to explore whether the brain activation pattern and connectivity patterns associated with T1 sentence and word processing could predict T2 language abilities. Because only 26 children finished the working memory task while undergoing fNIRS scanning and subsequently took part in the T2 assessments, the neural data from the working memory task was excluded from this analysis. We performed a correlation analysis between T1 working memory and T2 language comprehension. The results showed no significant correlation (r = 0.269, p = 0.174), suggesting the limited predictive power of T1 working memory for T2 language outcomes. SVR is a machine learning algorithm designed to train a model that can predict continuous target values based on input features. In this study, we applied SVR to develop separate predictive models for each of the three sentence conditions (i.e., SVO, Passive, SRC) and the word condition. These models were constructed using T1 neural activity and functional connectivity data as input features to predict T2 sentence comprehension and vocabulary scores.
For predictive models using activity levels (beta values) as input features, we constructed models based on activity patterns within four brain regions: left IFG, right IFG, left STG, and right STG. Each model was created for each of the four task conditions: SVO, Passive, SRC, and word. This approach resulted in a total of 16 models for each of the three dependent variables: T2 sentence comprehension, expressive vocabulary, and receptive vocabulary. Specifically, the beta values of each region for each task condition, combined with the T1 language assessment scores, served as input variables. The corresponding T2 language assessment scores served as the output variable.
In a similar manner, we constructed predictive models based on connectivity patterns. We developed three sets of connectivity models for each task condition, focusing on connectivity: (1) between the left and right IFG, (2) between the left and right STG, and (3) between the left IFG and left STG. This resulted in a total of 12 predictive models for each of the three dependent variables. In these models, functional connectivity patterns for each task condition, along with the T1 language assessment scores, were used as input variables, with the corresponding T2 language assessment scores as the output variable.
We employed the Leave-One-Out Cross-Validation (LOOCV) method for predictive analysis of the dataset. During each iteration, N-1 subsets of the data (training sets) were utilized for feature selection and model training, and the trained model was applied to the remaining subset (testing set) for prediction. This process iterated N times, providing predictions for each of the N subsets in the dataset. We also embedded a LOOCV within the model training process. The embedded LOOCV procedure mirrors the outer process. The regularization parameter 'c' was optimized within a [0,3] range, using a 0.1 step size, based on the mean absolute error (MSE) between predicted and actual scores. After model training, we conducted Pearson correlation analyses to assess the correlation between predicted and actual T2 scores and evaluated model performance using MSE and the coefficient of determination (R²). Significance was determined through permutation testing (1000 times), with correction for multiple models using the False Discovery Rate (FDR, q < 0.05). Further, we validated our findings using Relevance Vector Regression (RVR), which, like SVR, relies on a Bayesian framework but provides a sparser solution, thus reducing model complexity ( Tipping, 2001). This additional method helped confirm the reliability of our prediction models.
4 Result4.1 Behavioral tests
Table 1 shows the participants’ performance on behavioral measurements of sentence comprehension, vocabulary, and working memory, as well as behavioral performance during fNIRS tasks. We examined the skewness and kurtosis of the behavioral data, and it revealed an approximately normal distribution (-1.144 ≤ z's ≤ 1.922). The repeated-measures ANOVA on the ACC of the three sentence conditions that varied in syntactic complexity revealed a significant main effect [ F(2,48) = 38.455, p <0.001] ( Fig. 3 ). Post hoc comparison analysis (Bonferroni correction) revealed that the mean ACC for SVO sentence comprehension was higher than passive [ t(49) = 7.552, p < 0.001] and SRC sentences [ t(49) = 7.102, p < 0.001], but there was no significant difference between passive and SRC sentences [ t(49) = 1.902, p = 0.063]. Paired-sample t-test analysis of the mean ACC of 0-back and 1-back conditions in the working memory task revealed no significant difference between the two conditions [ t(39) = -1.689, p = 0.099].
Table 2 shows the correlations among behavioral measurements. Results revealed that SVO sentence comprehension was correlated with expressive vocabulary, whereas passive and SRC sentence comprehension were both correlated with expressive vocabulary and forward digit span. Moreover, SES was correlated with receptive vocabulary and sentence comprehension.
Table 3 shows the results of paired-sample t-tests comparing scores of sentence comprehension, vocabulary, and working memory at T1 and T2. The results indicated that the children exhibited statistically significant enhancements in all measured abilities at the subsequent measurement.
4.2 Brain activation and connectivity associated with syntactic processingWe first analyzed brain activation associated with sentence conditions by averaging the beta values of the three sentential conditions. Results showed that the sentence processing mainly activated the left IFG, MTG, STG, and right STG ( Fig. 4a ). We then compared brain activation when processing sentences of complex syntactic structures (passive and SRC sentences) to simpler structure (SVO sentences). The results showed that the passive and SRC conditions elicited greater activation mainly in the left IFG ( Fig. 4b).
Brain-behavior correlation analysis using linear regression showed that activation in the left IFG during sentence processing significantly correlated with behavioral performance of sentence comprehension even after controlling for children's age and gender, accounting for 18.2% of the variance in children's sentence comprehension performance ( t(46) = 2.298, R 2= 0.182, p = 0.026), as presented in Fig. 4c.
We further examined the connectivity patterns during sentence processing and tested whether regional connection differed among different syntactic conditions. The results of ANOVA analysis on FC strength showed significant differences among syntactic conditions between the left and right STG/MTG ( F(2,48) = 3.871, p = 0.024). Post hoc comparisons revealed that the FC strengths of the passive ( t(49) = 2.433, p = 0.019) and SRC condition ( t(49) = 2.259, p =0.028) were both stronger than the SVO condition. In addition, after controlling for gender and age, the strength of functional connectivity between the left and right STG/MTG during the passive sentence condition significantly correlated with passive sentence comprehension performance in the fNIRS task, accounting for 16.6% of the variance in children's sentence comprehension ( t(46) = 2.523, R 2 = 0.166, p = 0.015).
4.3 Relation between sentence processing, word processing, and working memory at the brain levelTo explore how the connections between syntactic development, vocabulary and working memory were instantiated in the brain, we conducted a correlation analysis between the activity of sentence condition and activation of word condition and working memory(1-back>0-back) in different ROIs. Results showed that brain activation in the left IFG during working memory task significantly correlated with brain activation during passive sentence processing ( r = 0.323, p = 0.04), but not with the activation during SVO ( r = 0.129, p = 0.427) or SRC ( r = 0.197, p = 0.222) sentence comprehension. On the other hand, brain activation during sentence processing was significantly correlated with word processing in the left STG/MTG ( r = 0.441, p = 0.005) and right STG ( r = 0.312, p = 0.049) ( Fig. 5 ). These results suggested that the correlation between syntactic processing during sentence comprehension and word comprehension may instantiate at different brain regions from those involved in working memory processes.
5 Predicting T2 performance based on neural data at T15.1 Predicted models based on activity patterns
The results showed that models based on T1 channel activity patterns as input features failed to predict T2 language abilities, as evidenced by the lack of significant correlations between the predicted scores and the actual language outcomes.
5.2 Predicted models based on connectivity patternsThe results of the predictive model based on connectivity patterns revealed that both sentence and word processing at T1 were predictive of T2 sentence comprehension, and T1 sentence comprehension predicted T2 vocabulary. Specifically, children's T2 sentence comprehension abilities could be predicted by functional connectivity between the left and right IFG during passive sentence processing ( r = 0.59, p < 0.001, MSE = 3.05, R² = 0.32) ( Fig. 6a ), as well as connectivity between the left and right STG/MTG during word processing ( r = 0.58, p < 0.001, MSE = 3.00, R² = 0.30) ( Fig. 6b). In addition, T2 receptive vocabulary was predicted by connectivity between the left IFG and left STM/MTG during SVO sentence processing ( r = 0.50, p = 0.0024, MSE = 1.82, R² = 0.23) ( Fig. 6c).
We also used RVR to validate predictive models, which yielded results that were consistent with the findings based on SVR, suggesting the reliability of our prediction models. Specifically, T2 sentence comprehension could be predicted using connectivity between the left and right IFG during passive sentence processing ( r = 0.60, p < 0.001, MSE = 3.36), as well as connectivity between the left and right STG/MTG during word processing ( r = 0.48, p = 0.004, MSE = 4.07). In addition, T2 receptive vocabulary was predicted by connectivity between the left IFG and left STM/MTG during SVO sentence processing ( r = 0.53, p = 0.0014, MSE = 1.78).
6 DiscussionIn this study, we employed fNIRS to explore the intricate relationships among syntax, semantics, and working memory in preschool-aged children, as well as how early neural patterns could potentially act as predictors of subsequent language development. Previous research has consistently highlighted the significant role of vocabulary and working memory in children's syntactic development ( Bates & Goodman, 1997; Moyle et al., 2007; Boyle et al., 2013; Verhagen et al., 2016). Our study not only corroborates the behavioral links between children's syntactic processing, vocabulary, and working memory but also elucidates the neural mechanisms that underpin these connections. We found that the activity level in the left IFG during the working memory task was correlated to that of processing complex sentences, indicating a link between working memory and syntactic processing in this region. Additionally, the relationship between word comprehension and sentence processing was observed in the posterior temporal regions, highlighting the role of these areas in integrating lexical and syntactic information. These findings contribute to a refined understanding of theoretical models of language development, demonstrating that it is not only influenced by individual cognitive factors but also by the intricate coordination among syntax, lexical-semantics, and working memory. Furthermore, by adopting a longitudinal study design and applying machine learning algorithms to the multivariate data, we have been able to establish predictive models of language development based on earlier brain connectivity patterns associated with sentence and word processing.
At the behavioral level, we found that children's ability to comprehend simpler sentences (i.e., SVO sentences) was significantly better than that of complex sentences (i.e., passive and SRC sentences). This finding aligns with prior research ( E. Kidd et al., 2016; He et al., 2020), suggesting that while preschool-aged children can comprehend sentences of relatively complex syntactic structure, their proficiency in understanding such sentences lags their grasp of simpler ones. Correlation analyses in our study indicated that children's expressive vocabulary scores positively correlate with their performance in both simple and complex (passive and SRC) sentence comprehension. On the other hand, their performance of forward digit span tasks is significantly related to the performance of processing complex sentences, but not with simpler SVO sentences. This finding aligns with previous studies' results ( Montgomery et al., 2008; Boyle et al., 2013). One possible explanation for this finding is that vocabulary may play a role in the fundamental abstract grammatical rules that are shared by both simple and complex sentences ( Bates & Goodman, 1997), whereas the working memory system may make a greater contribution to the temporary storage and processing of linguistic information in complex sentences.
At the brain level, previous studies have consistently demonstrated that sentence comprehension is supported by a frontotemporal network ( Friederici et al., 2017). We found that brain activation during sentence processing was distributed mainly in the left IFG, the left and right superior/middle temporal gyri. In addition, we found an effect of syntactic complexity in the left IFG in that it was particularly engaged during the comprehension of more structurally complex sentences. Furthermore, regression analysis revealed that the activation of this region could predict children's sentence comprehension performance even after controlling for age and gender. The model accounted for about 18% variance in children's sentence comprehension. Although this explains only a modest proportion of the variance, it underscores the significant influence of this neural activity on syntactic development in children. This finding adds to the growing body of evidence underscoring the importance of the left IFG as a neural scaffold supporting children's syntactic development ( Knoll et al., 2012).
Importantly, we found that the activity level in the left IFG during working memory task was correlated to that of processing complex passive sentences, but not with simpler ones. This finding furnishes a plausible explanation that the heightened engagement of the left IFG during the processing of syntactically complex sentences may be attributable to the need for more working memory resources. It is proposed that working memory is responsible for the temporary storage and manipulation of information ( Baddeley, 1998), which may constrain computational efficiency during sentence comprehension. Previous neuroimaging studies have found that neural overlap between sentence processing and working memory mainly occurs in the left IFG in adults ( Novick et al., 2010; Duncan and Owen, 2000), and this area is also sensitive to the syntactic movement processing that involved working memory during the sentence comprehension ( Santi and Grodzinsky, 2007). In addition, patients with focal damage to this region exhibit impairments in both language comprehension and working memory ( Novick et al., 2009).
During syntactic development, children gradually acquire the ability to process more complex sentence structures, such as relative clauses or passive sentences. For example, when children comprehend subjective relative clauses, they need to keep track of the sentence's subject to integrate with the coming verb filled by the object. Both working memory and syntactic processing tap into similar cognitive processes, reflecting the roles of encoding, storage, and manipulation inherent in working memory during sentence comprehension ( Lewis et al., 2006). Moreover, the development of syntactic ability and working memory in preschool children may be a mutually reinforcing process, whereby proficiency in storing and maintaining verbal input facilitates the improvement of language comprehension, and concurrently, the ability to maintain linguistic information promoting the development of working memory ( Jones et al., 2015; Schwering et al., 2020). Earlier research corroborates this association, indicating that children with language disorders tend to exhibit notably lower working memory capacities compared to their typically developing counterparts. This deficit is deeply intertwined with their sentence comprehension abilities ( Montgomery et al., 2018; Delage et al., 2020).
However, it is important to consider an alternative explanation that the observed correlation in the left IFG might reflect a shared phonological component that underlies both verbal working memory and complex syntactic comprehension. The left IFG is well-documented for its role in processing phonological information ( Boets et al., 2013; Price, 2012; Tan et al., 2005), which is crucial for both rehearsing verbal information in working memory and parsing complex syntactic structures during sentence comprehension. This dual role suggests that the correlation observed in our study between left IFG activation and syntactic processing may not be solely driven by working memory demands but could also be linked to the phonological processing necessary for maintaining and manipulating linguistic elements. Nonetheless, our study did not explicitly test whether this correlation extends to nonverbal working memory tasks. Future research that differentiates between verbal and nonverbal working memory could provide deeper insights into the nature of the correlation between syntactic processing and working memory, helping to clarify whether the observed relationship is primarily rooted in phonological processing or represents a more general working memory mechanism.
The correlation between word and sentence comprehension is mainly reflected in the left and right temporal regions. The temporal regions, particularly the middle and superior temporal regions, are well-established as key areas involved in lexical-semantic processing ( Binder et al., 2009; Price, 2012). For example, previous research found that the neural interaction between syntax and semantics was reflected in the left posterior superior temporal gyrus ( Skeide et al., 2014; Wang et al., 2020). These studies proposed that when the brain regions specialized for syntactic processing have not yet fully developed in preschool children, they may rely on the brain regions specialized for semantic processing, such as the posterior STG, to process syntax. The temporal cortex is suggested to be responsible for the integration and processing of both semantic and syntactic information in language comprehension ( Den Ouden et al., 2012). Our study found a significant association in neural activity between word processing and sentence processing in the left and right temporal cortex, suggesting that lexical-semantic processing in these regions may facilitate not only the retrieval of individual lexical items but also their integration with syntactic structures to construct meaningful sentences.
Moreover, we found that the functional connectivity between the left and right STG/MTG is particularly strengthened during the processing of syntactically complex structures (passive and SRC) than for simpler sentences (SVO). This enhanced connectivity was also correlated with children's performance on sentence comprehension tasks, indicating that stronger interhemispheric temporal connectivity was associated with better language skills ( Qi et al., 2021). These findings suggest that during early language development, when the left-lateralized network typically associated with language processing is not yet fully matured, children may rely on synchronized activity across both hemispheres to process complex sentences. Given that complex sentences often require more effort to process both syntactic and contextual information, the right STG/MTG's role in prosodic and contextual interpretation becomes particularly important ( Vigneau et al., 2011). Prosodic cues and contextual information help disambiguate complex sentence structures and clarify the relationships between sentence elements. The right STG/MTG's involvement may complement the left hemisphere's role in lexical access and semantic processing ( Binder et al., 2009; Price, 2012), making interhemispheric cooperation essential for integrating lexical-semantic and contextual information when processing syntactically complex sentences.
Crucially, our longitudinal investigation has yielded predictive models for language development based on earlier brain connectivity patterns associated with sentence and word processing. The results that T2 receptive vocabulary could be predicted using connectivity between the left IFG and left STM/MTG during sentence processing suggest an influence of earlier sentence comprehension on later vocabulary development. Furthermore, sentence comprehension at T2 can be predicted using functional connectivity between the left and right IFG during the passive sentences processing, and between the left and right STG/MTG during lexical processing at T1. The critical mass hypothesis proposed by Bates and colleagues underscores the importance of vocabulary in the acquisition and development of syntax ( Marchman & Bates, 1994; Bates & Goodman, 1997). Conversely, theories such as the syntactic bootstrapping hypothesis argue that an understanding of syntax can also catalyze the expansion of vocabulary ( Landau & Gleitman, 1985; Naigles, 1990). Our results align with a bidirectional perspective, supporting the interdependence of vocabulary and syntax as proposed by recent studies advocating for a bidimensional model of language development ( Pérez-Leroux et al., 2012). Such findings underscore the complex, reciprocal relationships between syntax and vocabulary growth, suggesting that enhancements in one can bolster the other.
In addition, the finding highlights the importance of inter-hemispheric communication as a robust predictor of subsequent sentence comprehension. Prior studies have consistently reported greater engagement of the right-hemispheric regions in children's language processing compared to adults ( Berl et al., 2014; Holland et al., 2007; Prat et al., 2023; Sharma et al., 2021; Szaflarski et al., 2006; Turkeltaub et al., 2003). It has been posited that the right hemisphere may serve a supportive function in the integration of varied linguistic information, especially when the left hemisphere's capabilities are still maturing or are less developed ( Lindell, 2006; Vigneau et al., 2011).
The right-hemispheric frontal and temporal regions play distinct yet complementary roles in language processing. A meta-analysis by Vigneau et al. (2011) highlights that the right IFG is involved in executive functions essential for language processing. These functions, though not tied to any single language component, are vital for managing and integrating complex linguistic information, especially in the context of complex sentence structures ( Hagoort, 2019). Additionally, research suggests that during language processing, children may rely more heavily on nonverbal cognitive functions, such as attention and inhibition ( Donnelly et al., 2011; Sharma et al., 2021). The connectivity between the left and right IFG observed in our study likely facilitates comprehensive sentence processing by integrating syntactic structures with these broader cognitive resources. This interhemispheric interaction may be particularly critical during early language acquisition when children are still mastering the complexities of syntax and semantics. Moreover, the connectivity between the left and right STG/MTG during lexical processing suggests a coordinated effort to integrate lexical-semantic information. This interhemispheric connectivity may enhance the ability to recognize words and interpret their meanings within a broader context, thereby supporting the development of sentence comprehension.
Our findings highlight the significant role of the right hemisphere in language development and emphasize the importance of interhemispheric interactions. Further investigations are necessary to determine if these predictive connectivity patterns persist in later developmental phases, given that children may show a natural progression toward less reliance on the right hemisphere and an increased specialization within the left hemisphere as their language functions mature. It is noteworthy that recent research shows that interhemispheric functional connectivity is a predictor of success in new language learning in adults ( Sander et al., 2023), indicating that inter-hemispheric communication is crucial not just in early language development but also plays a significant role in language acquisition during adulthood.
Finally, some limitations of this study are worth mentioning. Firstly, the spatial coverage of the fNIRS channels mainly encompasses the prefrontal cortex and a portion of the temporal cortex. This restricted coverage limits our ability to scrutinize activation and connectivity patterns across a broader range of brain regions. It is possible that we might overlook some potentially significant neural systems that lie outside the covered areas. For instance, the inferior parietal cortex has been identified in previous research as a crucial area for storing phonological information. It can be a candidate region for underpinning the connection between syntactic processing and working memory. Future research should consider employing more advanced imaging technologies or methodologies to circumvent this limitation. Secondly, our participant group was confined to children aged 4–6. Given that children's language and cognitive development are intricate, continuous processes, our study's age range may limit the generalizability of our findings to other age groups. Future studies should consider including a broader age range and adopting a longitudinal design to capture a more comprehensive picture of these developmental trajectories and causal relationships. Additionally, the sample's SES was reported to be higher than the broader city population. Since SES is correlated with vocabulary, syntax, and cognitive skills, this disparity may affect the generalizability of our results. More research is needed to explore how these findings might vary across different SES groups. Finally, the small sample size of children who participated in both the T1 working memory task and the T2 assessment constrained our ability to analyze their relationship effectively. This limitation underscores the necessity for further research with a larger cohort to better understand the neural underpinnings of the connection between working memory and syntactic development.
7 ConclusionsOur study provides evidence for the mechanisms for how the brain supports syntactic development in preschool children. The results highlight the critical role of the left IFG and bilateral temporal regions in children's syntactic development, and these regions may bridge connections with working memory and lexical-semantic processing during sentence comprehension. Processing syntactically complex sentences showed greater activity in left IFG and stronger functional connectivity between the left and right STG/MTG. We also found that the activity level in the left IFG during the working memory task was correlated to that of processing complex sentences, but not with simpler ones, suggesting that the heightened engagement of the left IFG during the processing of syntactically complex sentences may be attributable to the need for more working memory resources. On the other hand, the correlation between word and sentence comprehension is mainly reflected in the left and right temporal regions, suggesting that lexical-semantic processing in these regions may play a crucial role in sentence processing by facilitating the retrieval of lexical information. Furthermore, predictive models based on the inter-regional and inter-hemispheric connectivity of IFG and STG/MTG could successfully predict both sentence comprehension and vocabulary one year later, highlighting the importance of effective neural integration in facilitating language development.
DATA AVAILABILITY STATEMENT
The datasets and codes in this study are available from the corresponding author upon reasonable request. Requests for data access should be accompanied by a detailed project outline and are subject to a formal data sharing agreement to ensure compliance with data protection regulations.
CRediT authorship contribution statementDongsu Yan: Writing – review & editing, Writing – original draft, Visualization, Validation, Project administration, Methodology, Investigation, Formal analysis, Conceptualization. Tongfu Fang: Project administration, Methodology, Investigation, Conceptualization. Wei He: Writing – review & editing, Project administration, Methodology, Investigation, Conceptualization. Min Xu: Writing – review & editing, Writing – original draft, Supervision, Project administration, Methodology, Investigation, Funding acquisition, Conceptualization.
Declaration of competing interestThe authors declare that they have no conflict of interest.
AcknowledgementsWe thank children and parents who participated in this study. This work was supported by the National Natural Science Foundation of China ( 32171054), Shenzhen Innovation in Science and Technology Foundation for The Excellent Youth Scholars ( RCYX20210706092043066), Shenzhen-Hong Kong Institute of Brain Science-Shenzhen Fundamental Research Institutions (2023SHIBS0003).
| Measure | Mean (SD) | Range |
| Age (in months) | 61.5 (6.1) | 51∼75 |
| | ||
| Sentence comprehension | 30.0 (3.0) | 23∼35 |
| Receptive vocabulary | 26.9 (2.3) | 20∼31 |
| Expressive vocabulary | 18.1 (2.8) | 10∼23 |
| Forward digit span | 62.1 (19.5) | 28∼115 |
| Backward digit span | 13.2 (6.4) | 5∼29 |
| | ||
| SVO sentence | 88 (10.8) | |
| Passive sentence | 66.5 (19.9) | |
| SRC sentence | 71.3 (18.6) | |
| Word comprehension | 95.4 (6.6) | |
| 1-back WM | 94.3 (4.1) | |
| 0-back WM | 93 (5.9) |
| Measure | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 |
| 1.Age | - | |||||||||
| 2.SVO | 0.339 ⁎ | - | ||||||||
| 3.Passive | 0.210 | 0.284 ⁎ | - | |||||||
| 4.SRC | 0.349 ⁎⁎ | 0.475 ⁎⁎ | 0.582 ⁎⁎ | - | ||||||
| 5.Sentence comprehension | 0.277 | 0.285 ⁎ | 0.230 | 0.252 | - | |||||
| 6.Receptive | 0.156 | 0.122 | 0.335 ⁎ | 0.181 | 0.253 | - | ||||
| 7.Expressive | 0.211 ⁎ | 0.290 ⁎ | 0.476 ⁎⁎ | 0.343 ⁎ | 0.342 ⁎ | 0.51 ⁎⁎⁎ | - | |||
| 8.Forward digit span | 0.465 ⁎⁎ | 0.205 | 0.459 ⁎⁎ | 0.324 ⁎ | 0.330 ⁎ | 0.493 ⁎⁎⁎ | 0.481 ⁎⁎ | - | ||
| 9.Backward digit span | 0.417 ⁎⁎ | 0.146 | 0.221 | 0.141 | 0.205 | -0.016 | 0.003 | 0.040 | - | |
| 10.SES | 0.139 | 0.08 | 0.062 | -0.109 | 0.364 ⁎ | 0.409 ⁎ | 0.269 | 0.300 | -0.310 | - |
| Measure | T1 M(SD) | T2 M(SD) | T2 vs T 1 |
| Age (in months) | 58.2(2.8) | 70.2(2.8) | |
| | |||
| Sentence comprehension | 20.7(2.9) | 31.3(2.2) | |
| Receptive vocabulary | 26.7(2.3) | 28.62(1.6) | |
| Expressive vocabulary | 17.8(2.8) | 19.8(1.9) | |
| Forward digit span | 57.7(16.5) | 85.0(21.8) | |
| Backward digit span | 11.2(5.0) | 14.9(8.5) | |
©2024. The Authors