EEG N1 Specialization to Print in Chinese Primary School Students: Developmental Trajectories, Longitudinal Changes, and Individual Differences

ABSTRACT

Neural specialization for print can be indexed by the left‐lateralized N1 response as a tuning gradient to visual words, indicated by sensitivity (character vs. visual control) and selectivity (character vs. character‐like stimuli). Forty‐five Chinese children (20 boys) were recorded with EEG twice with a 2‐year interval during a character decision task (T1, 2016‐2017: 7–9 years old; T2, 2018‐2020: 9–11). Character N1 amplitude decreased faster with age (7–11 years) compared to non‐character N1, and character and character‐like N1 became less right‐lateralized. T1 better readers showed more longitudinal decrease of print sensitivity and more left‐lateralized T2 print sensitivity and selectivity. To conclude, reading skill drives functional neural efficiency for processing print, and the left hemisphere may be a linguistically universal neural mechanism for reading development.

Reading is one of the most important skills that children need to master in their primary school years. Brain areas in the left ventral occipital cortex (VOT) become increasingly specialized to process visual words during reading acquisition (Dehaene‐Lambertz et al. ²⁰¹⁸; Kubota et al. ²⁰¹⁹; Nordt et al. ²⁰²¹). Neural specialization to print is defined in two ways in the literature by different comparison conditions (Johnson 2011). Print sensitivity is defined by comparing it to a general or unrelated baseline condition, such as words versus a fixation cross (Ben‐Shachar et al. ²⁰¹¹), or words versus lower‐level visual control (Brem et al. 2009). In contrast, print selectivity is defined by the comparison of two closely related stimuli or conditions, e.g., words vs. drawing of verbally decodable objects (Centanni et al. 2017), words vs. word‐like stimuli, e.g., consonant strings that do not form words (Pegado et al. 2014). The current study refers to print sensitivity as the contrast between print and lower visual control and selectivity as the contrast between print and print‐like stimuli. It is important to note that print sensitivity in this study is also termed coarse tuning (e.g., Maurer et al. ²⁰⁰⁶), and print selectivity as fine tuning (e.g., Zhao et al. ²⁰¹⁴) in previous EEG studies. Although our definition of selectivity is within the print domain, evidence of print selectivity from visual objects in other categories is also reviewed, as it informs the development of neural specialization in children.

The neural specialization for print is also reflected in the electrophysiological responses generated by the VOT. There is an enhanced negativity (N1, also N170 in adults) in the left hemisphere that shows sensitivity and selectivity to visual word forms in adult readers (McCandliss et al. ²⁰⁰³). How this N1 specialization for print comes into being developmentally is not entirely clear. The progression of reading skill seems to be a main driver. Previous cross‐sectional studies have shown that higher reading skill is associated with increased N1 print sensitivity (Eberhard‐Moscicka et al. ²⁰¹⁵; Pleisch et al. ^2019b) and selectivity (Zhao et al. 2014) in beginning readers. Longitudinal EEG studies examining how reading skill influences the development of neural specialization in both aspects collectively are few (See Chyl et al. 2021 for a review and Wang et al. ²⁰²⁴ for recent evidence). Evidence is even more scarce for the development of neural specialization beyond the beginning stage of reading acquisition in a non‐alphabetic orthography.

The present study investigated the development of EEG N1 specialization for print from grade 1 to grade 6 using an accelerated longitudinal design, which involves collecting longitudinal data in multiple age groups. We examined the age‐related changes in N1 elicited by real characters, character‐like stimuli, and lower‐level visual control stimuli. Additionally, we investigated the prediction from earlier reading skill to subsequent levels and changes of print sensitivity and selectivity. This study was conducted in a non‐alphabetical language, specifically Chinese, which features non‐transparent grapheme‐sound correspondences. Understanding the development of N1 print specialization for Chinese characters would be helpful not only for characterizing the specific effects of learning to read in Chinese on children's brains, but also for enhancing our understanding of the universality of neural specialization for print. Furthermore, the children involved in this study were in grade 1–6 across the two measurement points. Studying readers of a wide age range covering the entire elementary school period provides a more comprehensive picture of the development of neural specialization for print.

1. The Development of Neural Sensitivity and Selectivity to Print

The development of N1 print sensitivity exhibits an inverted U‐shaped trajectory across a broad age range. Print sensitivity, defined as the contrast between print and false font, is the strongest in beginning readers who have received one to two years of formal literacy instruction, while it is weaker in pre‐readers and proficient readers (Fraga‐González et al. ²⁰²¹; Maurer et al. ²⁰⁰⁶). A similar developmental pattern was observed in a longitudinal fMRI study following ten children from preschool to primary school over six to seven MRI sessions (Dehaene‐Lambertz et al. ²⁰¹⁸). The researchers found that an enhanced activation to words in the visual word form area (embedded in VOT) had peaked at the onset of formal literacy instruction, which was then followed by a slight decline thereafter. For Chinese print, although the inverted‐U‐shaped curve has not been directly demonstrated, several studies have found that N1 print sensitivity declined from early to middle elementary school grades (e.g., Tong et al. ²⁰¹⁶). Thus, the inverted‐U‐shape can be assumed for the development of Chinese N1 print sensitivity, as other studies have shown that it did not present in Chinese kindergarteners who had no reading skill yet (Li et al. 2013).

The developmental trajectory of N1 print selectivity remains unclear. Some researchers have found that print selectivity, indexed by the contrast between print and print‐like stimuli, cannot be reliably detected in beginning readers (Wang et al. 2022; Zhao et al. ²⁰¹⁴) and emerges late in the late primary school years (Coch and Meade 2016). A recent study employing a novel paradigm, specifically Steady‐State Visual Evoked Potentials, demonstrated print‐selective responses (comparing well‐formed words to letter strings) in children in early elementary school years (K‐G3), and such selective responses increased greatly over a two‐year period (Wang et al. 2024). Development of print selectivity has also been investigated in fMRI studies. Centanni et al. (2017) found that the selectivity (between words and pronounceable drawings) was only present in adults, but not in grade 2–5 children, consistent with the late emergence of N1 selectivity. Additionally, Nordt et al. (2021) observed that visual word selectivity, compared to other visual objects such as faces, places, and limbs, increased in a linear fashion from ages 5 to 17 in the left ventral temporal cortex.

Regarding the lateralization of the neural reading network, results from longitudinal fMRI studies suggest that the left hemisphere becomes increasingly involved in processing print as development progresses (Dehaene‐Lambertz et al. ²⁰¹⁸; Nordt et al. ²⁰²¹). In contrast, the literature on EEG studies is less consistent regarding the lateralization of print sensitivity and selectivity. Maurer et al. (2005) found that prereaders with high letter knowledge exhibited N1 sensitivity to letter strings (compared against symbol strings), and the effect was only observed in the right hemisphere. This early right‐lateralized print sensitivity may be driven by mere visual exposure not reading (Posner and McCandliss ¹⁹⁹⁹). Conversely, using a different EEG approach (frequency‐tagging), Lochy et al. (2016) demonstrated that prereaders exhibited left‐lateralized sensitivity for letter strings among pseudofonts. After formal literacy instruction begins, in grade 2, Maurer et al. (2011) found left‐lateralized N1 to real words in contrast to the right‐lateralized N1 to symbol strings. On the other hand, using the same conditional contrast, Eberhard‐Moscicka et al. (2015) reported bilateral N1 to words and symbols in Swiss German grade 1 students, resulting in the bilateral N1 print sensitivity. Similarly, a study by Spironelli and Angrilli (2009) also found that 10‐yearyear‐old children exhibited bilateral N1 to real words in contrast to fixation marks. For N1 print selectivity, Zhao et al. (2014) identified a bilateral distribution for both word and consonant‐string N1; however, the contrast was only significant in the left hemisphere in advanced readers. The inconsistent findings regarding lateralization may be attributed to writing systems, study design, analysis approach, and developmental stage. A consistent developmental pattern emerges: older and more advanced readers demonstrate greater left‐lateralization of N1 print specialization in alphabetic contexts (Maurer et al. 2005, 2011; Spironelli and Angrilli ²⁰⁰⁹; Zhao et al. ²⁰¹⁴), despite inconsistent outcomes. This developmental pattern is corroborated by a recent longitudinal study that showed the EEG print selectivity responses (words vs. letter strings) became increasingly left‐lateralized over time in elementary school children (Wang et al. 2024).

The left‐lateralization with development is primarily driven by the audiovisual integration of speech and visual words (Maurer and McCandliss ²⁰⁰⁷; Varga et al. ²⁰²⁰). In alphabetic writing systems, the dominant linguistic structural constituent that is mapped onto print is phonology, which is processed in the left temporooccipital cortex. Activation in areas related to phonological and audiovisual integration is linked to the development of reading skill (see Chyl et al. ²⁰²¹ for a review). In contrast to alphabetic scripts, Chinese script is notorious for its opaque mapping between phonology and orthographic form. The orthographic representation of individual phonemes does not exist in Chinese, and the orthographic cues for syllables are very unreliable. Consequently, phonological skills play a limited role in reading acquisition in Chinese. Instead, morphological and orthographic skills are critical and account for a significant amount of individual variation in word reading outcomes in children (McBride ²⁰¹⁶). These print‐specific characteristics render the lateralization of neural specialization for Chinese print a debatable issue. There are two perspectives. The first emphasizes the additional engagement of the right hemisphere in processing Chinese print, based on the bi‐lateralized or right‐lateralized Chinese print effect observed in BOLD signals in adults (Perfetti et al. 2013; Tian et al. ²⁰²⁰). The involvement of the right hemisphere in processing Chinese characters can be attributed to its advantage in holistic processing (Wong et al. 2012) or in processing low spatial frequency information of Chinese characters, such as the spatial relationship between radicals (Chung et al. 2018).

The alternative perspective emphasizes the role of the left hemisphere. The neural convergence between speech and print is linguistically universal and occurs dominantly in the left hemisphere (Rueckl et al. 2015). For example, the left superior temporal gyrus (STG), a core area in the reading network across different writing systems, helps segment core structural constituents of a given language. This includes phonology in alphabetic languages (Maurer and McCandliss ²⁰⁰⁷) and morphology in Chinese (Rueckl et al. 2015; Zhang et al. ²⁰²³). Supporting this view, several cross‐sectional N1 ERP studies conducted with Chinese children have demonstrated an increasingly important role of the left hemisphere in processing Chinese characters as they develop. Li et al. (2013) found left‐lateralized character N1 emerged early in grade 1 and kindergarten children, in contrast to the bilateralized N1 elicited by line drawings of houses and faces. Tong et al. (2016) found that young Chinese readers (grade 1–2) showed bilateralized print sensitivity (in contrast to scrambled characters) and selectivity (in contrast to illegal characters), whereas older readers (grade 3–4) showed left‐lateralization.

2. The Longitudinal Relation Between N1 Neural Specialization and Reading Skill

Neural specialization for print is largely driven by reading experience and improvement of reading skill rather than by general maturation, as stated in the reading expertise hypothesis (McCandliss et al. ²⁰⁰³). For N1 print sensitivity, evidence from training studies is substantial that explicit teaching of phoneme‐grapheme corresponding rules enhances N1 responses to the taught script (Brem et al. 2010; Pleisch et al. ^2019a; Yoncheva et al. ²⁰¹⁵). Several longitudinal studies have found a correlation between N1 neural specialization for print and reading skill. For example, the longitudinal change of print selectivity strength is correlated with gains in rapid naming (Wang et al. 2024). Earlier print sensitivity has been shown to predict subsequent reading outcomes (Eberhard‐Moscicka et al. ²⁰²¹). However, the predictive relation between the lateralization and reading skill is less documented, although the process of left lateralization of print sensitivity (Maurer et al. 2011) and selectivity (Wang et al. 2024) cooccurs with reading skill development. Based on previous evidence, it is plausible that individual differences in the developmental trajectories of N1 print specialization can be explained by variations in reading skills. Maurer et al. (2011) found that word N1 amplitude decreased from grade 2 to 5 in typical readers but remained stable in children with dyslexia (extremely low performance in reading fluency). No studies have investigated individual differences in the development of N1 print specialization in children with a continuous reading skill distribution.

Considering the research gap, the present study had two aims. The first aim was to examine the developmental trajectories of N1 to print and non‐print stimuli, i.e., Chinese characters, character‐like stimuli, low‐level visual controls, throughout the elementary school years. Previous studies have demonstrated that reading experience (development) not only influences the neural processing of print but also affects print‐like stimuli as well as visual objects of other categories (Dehaene‐Lambertz et al. ²⁰¹⁸; Nordt et al. ²⁰²¹). In EEG studies, reading and reading‐related cognitive‐linguistic skills are associated with the N1 amplitude elicited by real words, well‐formed pseudowords, consonant strings, and false fonts in elementary school children (Coch and Meade 2016). Therefore, we predicted that the N1 changes with age in all three conditions, but at different rates reflecting print specialization. We do not have specific hypotheses regarding the developmental trajectory for each condition, as the evidence is inconsistent.

The second aim was to investigate the group‐level longitudinal change and individual variations in N1 neural specialization, as indexed by print sensitivity and selectivity. Examining both aspects allows for testing the interactive specialization (IS) theory of brain function development, which posits that neural sensitivity develops prior to and lays a foundation for the development of neural selectivity (Johnson 2011). Previous findings on neural specialization for print provide evidence supporting this theory, suggesting that print sensitivity and selectivity follow distinct developmental trajectories. Print sensitivity decreases after the beginning stage of learning to read, and print selectivity emerges later and increases with development. Regarding lateralization, the IS theory predicts that print sensitivity and selectivity become more localized with development. In the case of visual word recognition, research evidence indicates a consistent developmental pattern that print sensitivity and selectivity become more left‐lateralized over time. Therefore, we predicted that (1) print sensitivity would decrease while print selectivity would increase across time, and (2) print sensitivity and selectivity would become more left‐lateralized with time.

Regarding sources of individual variation, we chose to test the skill of word reading fluency. In Hong Kong, formal literacy instruction starts in the first year of kindergarten when children are approximately 3 years old. By the start of grade 1, children have had 3 years of literacy experience and thus have developed a certain level of word‐reading fluency, which continues to develop across primary school years. Informed by the reading expertise hypothesis (McCandliss et al. ²⁰⁰³), we predicted that reading skill at T1 would explain the individual variations in T2 print neural specialization measures, as well as the longitudinal changes from T1 to T2. Particularly, more fluent readers at T1 would exhibit greater left‐lateralization at T2 for print sensitivity and selectivity and more enhanced print selectivity at T2. T1 fluent readers would show more decrease (or less increase) in print sensitivity and more increase in print selectivity.

3. Methods

3.1. Participants

The participants were from a large‐scale longitudinal twin study in Hong Kong (Wong et al. 2017). 50 children from 25 twin pairs from grades 1 to 4 were recorded EEG signals twice with an interval of 2 years. Details regarding participant selection are provided in the Supporting Information. We selected participants who had good EEG data quality (criteria are described in the EEG Preprocessing section below) at both times of data collection. 45 participants (20 boys) from 25 twin pairs (MZ = 12) were selected. The mean age was 96 months at the first time of EEG. The participant characteristics are in Table 1. They were all Chinese children whose native language was Cantonese. They all had nonverbal IQ scores > 80, normal or corrected‐to‐normal visual acuity, and no neurological or cognitive impairment according to teacher and parental report. Data at the first time of recording was a subset of the data published earlier (Lo et al. 2019). The first wave of data collection was conducted from September 2016 to Dec 2017, and the second wave from September 2018 to January 2020. The average interval between two waves of EEG assessment was 2.04 (SD = 0.13) years; the reading assessment was conducted on average 2 months (SD = 1.2) after the first wave of EEG data collection. The reading outcomes and related skills at both time points and the longitudinal increase are reported in Supporting Information (Figure 1).

TABLE 1.

Sample characteristics.

	Grade* 1 (n = 3)	Grade 2 (n = 25)	Grade 3 (n = 13)	Grade 4 (n = 4)	Total (N = 45)
Gender (male)	2	14	1	3	20
T1 age	7.17 (0.21)	7.58 (0.24)	8.41 (0.41)	9.11 (0.04)	7.92 (0.62)
T2 age	9.19 (0.14)	9.60 (0.33)	10.48 (0.49)	11.20 (0.13)	9.97 (0.68)
T1 WRF_z	0.79 (0.34)	−0.01 (0.84)	−0.23 (1.11)	1.21 (0.69)	0.22 (0.89)
T1 WRF_raw	57.33 (6.02)	56.60 (15.84)	62.38 (21.73)	97.50 (13.69)	61.96 (20.32)

Abbreviation: WRF, word reading fluency.

^*Grade level at T1.

FIGURE 1.

Experimental procedures and global field power for each experimental condition across two time points.

3.2. EEG Experimental Procedure and Materials

Children needed to perform the task of character decision while their EEG signals were recorded. The task originally comprised four conditions, i.e., real characters, pseudo characters, illegal characters, and visual control stimuli, each containing 60 trials (Authors, 2019). Real characters were all compound characters with the left–right structure. That is, the real Chinese characters comprised one semantic radical and one phonetic radical placed horizontally, with the phonetic radical always on the right. All real characters used in this study were selected from a curriculum list for elementary school children published by the Chinese Language Education Section of Hong Kong (2009). The radicals in the real characters were used to form pseudo and illegal characters. Pseudo characters were combinations of real radicals, which conformed to orthographic regularities but did not exist in the Chinese writing system. Illegal characters were combinations of semantic and phonetic radicals in illegal positions, created by reversing the radicals in real Chinese compound characters. The visual control stimuli were adopted from a previous study (Su et al. 2015) and constructed by randomly combining strokes. Stroke numbers of the stimuli were the same across conditions to control for visual complexity. For the current study, the pseudo character condition was removed from the analysis due to a low accuracy rate. Examples of the remaining three conditions and example trials are shown in Figure 1 A.

Participants were asked to indicate whether a stimulus was a real character or not by pressing the corresponding button on a serial response (SR) box. For each trial, a fixation mark “+” was shown on the screen for 500 ms. It was then replaced by a stimulus with a presentation duration of 1500 ms. After the stimulus presentation, a blank screen was shown for 1000 ms as an interval. All stimuli were presented in a font size of 56 to children who were seated 80 cm away from the computer monitor. Each stimulus covered 0.82^ο of both vertical and horizontal visual angles. The stimuli were presented randomly to children. We used E‐Prime software to present the stimuli and record children's accuracy and reaction times.

The EEG recordings were conducted in a sound‐attenuated EEG lab at the university. Before the experiment, a research assistant introduced the procedure to the child. EEG was recorded using the HydroCel GSN EGI 128‐channel system (EGI net station v5.3, Electrical Geodesics Inc., Eugene, Oregon). The sampling rate was 500 Hz, and electrode Cz was used as the online reference. The electrode impedance was kept below 50 kΩ. Based on a separate timing test, the time delay (12 ms) between the start of the actual stimulus and the event trigger was later corrected in data preprocessing.

3.3. Word Reading Fluency

Word reading fluency was measured by the task of 1 min Chinese word reading adapted from the 1‐Minute Word Reading subtest of the Hong Kong Test of Specific Learning Difficulties in Reading and Writing for Primary School Students—Second Edition (HKT‐P[II], Ho et al. ²⁰⁰⁷). Children were asked to read aloud 120 highly frequent Chinese two‐character words as quickly and accurately as possible within 1 min. The number of words read correctly within 1 min was the score. Standardized residuals of the raw scores regressed on age and gender in a representative sample of Hong Kong Chinese children (n = 2808) were used in the analysis.

3.4. EEG Preprocessing

The EEGLAB version 2021.0 was used to preprocess EEG data. The same processing pipeline (Authors, 2019) was adopted to process T1 and T2 data. The preprocessing steps included (1) down‐sampling to 250 Hz; (2) subdividing channels into 10–10 system montage and removing the unmapped 59 electrodes; (3) bandpass filtering at 0.3–30 Hz; (4) removing bad channels that had too much variance with PREPpipeline (for details please refer to Bigdely‐Shamlo et al. ²⁰¹⁵); (4) running independent component analysis using the infomax ICA algorithm of Bell and Sejnowski (1995), (5) removing eye‐movement‐related artifact components automatically using ADJUST (Mognon et al. 2011); (6) using spherical spline to interpolate removed channels in the 10–10 system montage; (7) segmenting epoch at –150 ms to 850 ms; (8) removing trials that exceeded the −80 to 80 mV thresholds; (9) select trials that had correct responses. This pipeline was selected because it yielded the best signal‐to‐noise ratio for the N1 ERP component.

We selected participants for data analysis based on the following criteria: (1) had a signal noise ratio (SNR) above 3 on the expected P1 or N1 ERP components (1 removed), (2) scored above 70% on the behavioral task (1 removed), and (3) had at least 10 trials in each condition (3 removed). The average number of trials in each condition at two time points was 49.4 (T1 visual control), 48.3 (T1 real character), 49.9 (T1 illegal character), 51.8 (T2 visual control), 50.3 (T2 real character), and 50.1 (T2 illegal character). Based on previous findings that the N1 component is predominantly detected at occipitotemporal sites (e.g., Maurer et al. ²⁰⁰⁸; Tong et al. ²⁰¹⁶), the electrodes of P7, PO7, P8, and PO8 (at Figure 2), the representative electrodes of the occipitotemporal sites, were selected for statistical analyses. The global field power (GFP) plot indicates that the N1 component at the second time point ended slightly earlier than t1 (Figure 1 B). The N1 time window was thus selected to be from 160 to 275 ms at T1 and from 160 to 270 ms at T2.

FIGURE 2.

Developmental trajectories of (A) N1 strength and (B) lateralization to different stimuli across elementary school years estimated by GAMM. The estimated trajectories stacked across conditions are in the upper left corners, and the lines for each condition with individual trajectories are plotted in the rest of the panels.

3.5. Data Analysis

To address the first aim of the study, generalized additive mixed models (GAMM) were employed to analyze N1 responses at the trial level. GAMMs are a type of statistical model that combines the flexibility of generalized additive models (GAMs) with the ability to account for random effects in mixed‐effect models. GAMMs can handle the possible non‐linear relation between age and ERPs as observed previously (e.g., Fraga‐González et al. ²⁰²¹), as well as data independence due to the hierarchical nature of the data. In the case of the current study, trials were nested in individuals which were further nested in families. One GAMM model was built for N1 amplitude and lateralization each. The lateralization index was computed as (P7 + PO7‐P8‐PO8)/ GFP; the normalization was to remove the influence of strength from distribution. The trial‐level analysis was adopted to take into account the intraindividual variance of the N1 response and to maximize the power of analysis. The GAMM models, as written in R code below, included condition and age as fixed effects. The smooth term of age was estimated by condition. Random effects included random intercept of family and subject, and random slopes of time by subject id. The data was analyzed using the mgcv package (Wood 2023) in the R platform.

$$\begin{gathered} \begin{array}{c}\text{N1 Strength}~\text{condition}+s\left(\mathrm{age} \mathrm{by}\text{condition}\right)\hfill \\ +\left(1+\text{Time}|\text{Subject}\right)+\left(1|\text{family}\right); \\ \text{N1 Lateralization}~\text{condition}\hfill \\ +s\left(\mathrm{age}\mathrm{by}\text{condition}\right)\hfill \\ +\left(1+\text{Time}|\text{Subject}\right)+\left(1|\text{family}\right).\hfill \end{array} \end{gathered}$$

To address the second aim of the study, linear mixed‐effect models (LMM) were adopted to examine the level and longitudinal change of conditional contrasts, i.e., sensitivity (real character‐visual control) and selectivity (real character—illegal character). Moreover, the random effects in LMM allow estimation of individual variations in the change and final level of N1 responses as well as those of the conditional contrasts. We can then further examine if T1 reading skill predicts T2 print sensitivity and selectivity and their longitudinal changes. T2 character responses were set as reference. The fixed effects included a three‐way interaction between time, condition, and T1 reading skill, and the embedded one‐way and two‐way contrasts. For the N1 strength model, random effects included random intercepts of subject, family, and electrodes, as well as random slopes of time, condition, and time × condition by subject to indicate inter‐individual variability of longitudinal change and conditional contrasts. The random effects were the same for the N1 lateralization model, except that they did not include a random intercept of electrodes or the random slopes of the time × condition interaction; the removal of the latter contributing to model convergence. The models written in R code are shown below.

$$\begin{gathered} \begin{array}{c}\text{N1 Strength}~\text{Time}\times \text{Condition}\times \mathrm{T}1\_\text{reading}\hfill \\ +\left(1+\text{Time}\times \text{Condition}|\text{Subject}\right)\hfill \\ +\left(1|\text{family}\right)+\left(1|\text{electrode}\right); \\ \text{N1 Lateralization}~\text{Time}\hfill \\ \times \text{Condition}\times \mathrm{T}1\_\text{reading}\hfill \\ +\left(1+\text{Time}|\text{Subject}\right)+\left(1|\text{family}\right).\hfill \end{array} \end{gathered}$$

The t tests were conducted to detect the statistical significance of the fixed effects. Post hoc pairwise comparisons were performed in light of significant group‐level interactions, i.e., time and condition. The p‐values were adjusted for multiple tests using the fdr method as implemented in the emmeans package (Lenth 2023). For modulation effects of reading, post hoc LMM analysis was conducted in light of significant contrasts and interactions, predicting the individual T2 level or growth in contrasts (print sensitivity or selectivity) using T1 reading. The p‐values of the post hoc LMMs were also fdr corrected. The LMM models were estimated in the ‘lmertest’ package (Kuznetsova et al. 2017) on the R platform. The output of random effects in the N1 strength and lateralization model is in the Table S3.

LMM was also employed to analyze the accuracy rate and response time (averaged across correct trials) in the lexical decision task at the condition level. Time, condition, and the interaction were set as the fixed effects. Family ID was set as a random intercept to account for the data dependence in twin pairs. In addition, the d‐prime index and criterion (c) of hitting the target (character), informed by the signal detection theory, were also analyzed with time as the fixed effect.

At last, we conducted exploratory analysis on the reliability and stability of measures of N1 print specialization, namely print sensitivity strength, print sensitivity lateralization, print selectivity strength, and print selectivity lateralization. Independent sample t statistics were calculated by comparing the trial‐level responses between character and non‐character conditions for each participant. To assess longitudinal stability, correlation coefficients were computed for the t statistics between T1 and T2. To assess their intraindividual stability (reliability), we randomly split the trials in half by condition for each individual and computed the t statistics for each half. Pearson correlation coefficients were computed for the t statistics between two halves.

4. Results

4.1. Behavioral Results: Character Decision Task

Response accuracy and time in the EEG task are shown in Table 2. The results of the LMM model on response accuracy showed significant main effects of time, F (1,220) = 57.01, p < 0.001, and condition, F (2,220) = 14.86, p < 0.001. The interaction between time and condition was significant, F (2,220) = 2.35, p = 0.001. Post hoc analysis was performed to show the condition differences at two time points. At T1, the accuracy in the illegal character condition was lower than that in the visual control condition (p _fdr < 0.05); the real character condition did not show differences from the illegal character and the visual control conditions (both p _fdr > 0.05). At T2, the real character was lower than the visual control condition (p _fdr < 0.05); the illegal character did not show differences from the other two conditions (both p _fdr > 0.05).

TABLE 2.

Descriptive statistics of behavioral performance in the EEG task.

	Accuracy		Response time
	Time 1	Time 2	Time 1	Time 2
Control	0.91 (0.12)	0.96 (0.08)	777.92 (120.18)	639.94 (101.37)
Real	0.85 (0.13)	0.89 (0.12)	829.81 (120.74)	704.65 (89.75)
Illegal	0.79 (0.12)	0.92 (0.09)	878.39 (156.65)	706.84 (113.26)

The results on response time showed significant main effects of time, F (1,220) =212.26, p < 0.01, and condition, F (2,220) = 6.40, p < 0.01. The interaction between time and condition was not significant, F (2,220) = 1.94, p > 0.05. The response time in the visual control condition was shorter than in the illegal character condition (p _fdr < 0.05). The real character, the illegal character, and the visual control conditions did not significantly differ from each other (p _fdr > 0.05). The overall response time decreased over time.

The results on d prime showed a significant effect of time, F (1,65) =17.21, p < 0.01. Children's performance detecting real characters in the lexical decision task significantly increased from T1 (Mean = 2.22; SD = 0.75) to T2 (Mean = 3.12; SD = 0.94). There was no significant time effect on the criterion (c) for detecting characters versus the distractors, F (1,65) = 0.22, p > 0.05; the performances were comparable between T1 (Mean = 0.26; SD = 0.26) and T2 (Mean = 0.26; SD = 0.30).

4.2. Developmental Trajectories of N1 Responses Elicited by Print and Non‐Print Stimuli

Results of the N1 strength GAMM showed that the N1 strength declined with age in a linear fashion in character (edf = 1.00, F = 21.26, p < 0.001) and illegal character conditions (edf = 1.00, F = 9.38, p = 0.002). N1 amplitude decrease for each 1‐year increase in age ranged from 0.72 to 1.74 for real characters (B = 1.22, t = 4.61, p < 0.001) and 0.31 to 1.34 for illegal characters (B = 0.82, t = 3.06, p = 0.002). In contrast, the N1 strength elicited by visual control stimuli showed a very non‐linear relation with age (edf = 2.56, F = 17.29, p < 0.001). The N1 declined until age 10 and then slightly increased from age 10 to 11. The developmental trajectories of N1 strength are shown in Figure 2.

Results of the N1 lateralization GAMM showed that the N1 became less lateralized with age in a linear fashion for the illegal character condition (edf = 1.00, F = 4.41, p = 0.036). The standardized lateralization index change for each 1‐year increase in age ranged from 0.01 to 0.24 for illegal characters (B = −0.13, p = 0.036). In contrast, the lateralization of character N1 showed a non‐linear relation with age (edf = 2.01, F = 4.75, p < 0.01). The N1 became less right‐lateralized from age 7 to age 9 and then remained stable until 11. At last, no relation was found between lateralization of visual control N1 and age (edf = 1.00, F = 0.11, p = 0.74) (Figure 2). The full output of the GAMM models is in the Table S2.

4.3. Longitudinal Change of N1 Print Specialization and Modulation of T1 Reading Skill

The fixed effects of the N1 strength LMM are shown in Table 2. The results showed that the two‐way interaction between time and print selectivity contrast was significant, B = 0.76, p < 0.05, indicating the strength of print selectivity decreased across time at the group level (Figure 3A). The two‐way interaction between print sensitivity and time, on the other hand, was not significant, p > 0.05. Post hoc comparisons showed that N1 in all conditions decreased significantly across time (all p _fdr < 0.01); print sensitivity and selectivity were significant at both time points (all p _fdr < 0.01) (Figure 3C). The significant interaction between time and print selectivity was driven by a larger longitudinal decrease of the real character N1 (B = 1.94, p _fdr < 0.01) than the illegal character N1 (B = 1.14, p _fdr < 0.05) (Figure 3C). The full output of pairwise post hoc contrasts is in the Table S4.

FIGURE 3.

EEG waves of raw N1 and conditional contrast strength (A) and lateralization (B) accompanied by topographic maps (E). Boxplots for N1 amplitude (C) and lateralization (D) and the significant group‐level contrasts estimated in the LMM models. *p _fdr < 0.05_; **p _fdr < 0.01; ***p _fdr < 0.001.

Regarding the modulating effect of reading skill, the three‐way interaction between time, print sensitivity contrast, and reading skill was significant, B = −1.06, p < 0.05, suggesting that reading skill modulated the longitudinal change of print sensitivity. To verify this interpretation, the post hoc LMM regression of individual print sensitivity growth (predicted from the random effect model) on T1 reading skill showed that the higher reading level was significantly linked to a greater decrease in print sensitivity, B = −0.89, p _fdr < 0.05 (Figure 4A).

FIGURE 4.

T1 reading skill predicted the (A) longitudinal change of print sensitivity strength, (B) T2 print sensitivity lateralization, and (C) T2 print selectivity lateralization. The beta regression coefficients in all three plots are significant (p _fdr < 0.05) The gray ribbons indicate confidence interval. Positive values in plot A indicate decrease of N1 sensitivity (negative). Children from the same family are represented with the same symbols. The intraclass correlation coefficients, are 0.35, 0.27, 0.37 for regression in plot a, b, and c, representing the variance explained by family clustering. The predictions still stand after age is controlled for (Supporting Information).

The fixed effects of the N1 lateralization LMM are shown in Table 3. The results showed a significant two‐way interaction between time and print sensitivity contrast, B = −0.18, p < 0.05, indicating that print sensitivity became less right‐lateralized across time at the group level (Figure 3B). The two‐way interaction between print selectivity and time, on the other hand, was not significant. Post hoc comparisons showed that print selectivity contrast was significant at T1, B = −0.22, p _fdr < 0.001, and T2, B = −0.19, p _fdr < 0.05, indicating that the print selectivity was left‐lateralized across time. The print sensitivity contrast, on the other hand, was not significant at either time point, indicating bilateralized print sensitivity (Figure 3D), although a pattern was suggested that print sensitivity was slightly right‐lateralized at T1 and left‐lateralized at T2 (Figure 3B). The real character N1 (B = −0.19, p _fdr < 0.001) and illegal character N1 (B = −0.16, p _fdr < 0.05) became significantly less right‐lateralized from T1 to T2, whereas the visual control N1 did not change across time (Figure 3D). The full output of pairwise comparison is shown in the Table S4 (Table 3).

TABLE 3.

The fixed effect contrasts in the LMM models for N1 strength and lateralization.

	N1 strength model			N1lateralization model
	Estimates	CI	p	Estimates	CI	p
(Intercept)	−5.83	−7.63 to −4.03	< 0.001	0.05	−0.17 to 0.27	0.668
read_T2c	−0.12	−1.39 to 1.15	0.850	−0.30	−0.52 to −0.08	0.007
LCc	1.90	0.93 to 2.87	< 0.001	0.14	‐0.08 to 0.36	0.200
T2sen	−5.20	4.34 to 6.06	< 0.001	−0.06	−0.06 to 0.18	0.343
T2sel	−1.67	1.11 to 2.22	< 0.001	−0.20	0.08 to 0.32	0.001
read_LCc	−0.43	−1.44 to 0.58	0.406	0.18	−0.04 to 0.41	0.113
read_T2sen	−0.14	−1.02 to 0.75	0.765	−0.30	−0.43 to −0.18	< 0.001
read_T2sel	0.01	−0.56 to 0.59	0.961	−0.20	−0.32 to −0.07	0.002
LCsen	−0.18	−1.08 to 0.73	0.703	−0.18	−0.35 to −0.00	0.048
LCsel	0.76	0.05 to 1.46	0.036	−0.01	−0.19 to 0.17	0.917
read_LCsen	1.13	0.19 to 2.08	0.019	−0.12	−0.31 to 0.06	0.194
read_LCsel	0.44	−0.30 to 1.18	0.243	−0.06	−0.25 to 0.12	0.517

Note: Intercept, T2 character N1; read_T2c, association between T1 reading skill and T2 character response; LCc, longitudinal change of character response (T2 real‐T1real); T2sen, print sensitivity at T2 (T2real‐T2visual control); T2sel, print selectivity at T2 (T2real‐T2illegal characters); read_LCc, association between T1 reading skill and longitudinal change of character response (read* [T2 real‐T1real]); read_T2sen, association between T1 reading skill and T2 print sensitivity (read* [T2 real‐T2 visual control]); read_T2sel, association between T1 reading skill and T2 print selectivity (read* [T2 real‐T2 illegal]); LCsen, longitudinal change of print sensitivity ([T2 real‐T2 visual control]‐[T1 real‐T1 visual control]); LCsel, longitudinal change of print selectivity ([T2 real‐T2 illegal]‐[T1 real‐T1 illegal]); read_LCsen, association between T1 reading skill and longitudinal change of print sensitivity (read*[[T2 real‐T2 visual control]‐[T1 real‐T1 visual control]]); read_LCsel, association between T1 reading skill and longitudinal change of print selectivity (read*[[T2 real‐T2 illegal]‐[T1 real‐T1 illegal]]).

Regarding the modulating effects of reading skill, the two‐way interaction between print sensitivity contrast and reading skill was significant, B = −0.30, p < 0.05, and so was the interaction between print selectivity and reading skill, B = −0.20, p < 0.05, indicating that higher T1 reading skill was linked to more left‐lateralized print sensitivity and selectivity at T2. To verify this interpretation, the t statistics for print sensitivity and selectivity contrasts were calculated for each participant at T2. LMMs were conducted predicting print sensitivity and selectivity using T1 reading score as a fixed effect and a random intercept of family. The results showed that higher T1 reading skill was significantly linked to more left‐lateralized print sensitivity (B = −0.66, p _fdr  = 0.04) and selectivity (B = −0.48, p _fdr = 0.04), as shown in Figure 4B,C. The three‐way interaction between print sensitivity contrast, time, and reading skill was not significant, nor was the three‐way interaction between print selectivity contrast, time, and reading skill (both ps > 0.05), suggesting that T1 reading skill did not explain the individual variation in the longitudinal changes of print sensitivity nor selectivity lateralization.

4.4. Exploratory Analyses

Table 4 shows the correlation coefficients of t statistics between halves of trials (reliability) and between T1 and T2 (stability). The reliability was high (all r > 0.62) for print sensitivity strength and lateralization at both time points, whereas it was low for print selectivity measures (all r < 0.48) except for lateralization at T2 (r = 0.68). All N1 print specialization measures showed significant longitudinal stability (p < 0.05) except for print selectivity strength.

TABLE 4.

The reliability and stability of N1 print sensitivity and selectivity measures.

		Reliability		Longitudinal stability
		T1	T2	T1 and T2
Print sensitivity	Strength	0.68 [0.49, 0.81]	0.75 [0.59, 0.85]	0.47 [0.35, 0.58]
	Lateralization	0.63 [0.37, 0.76]	0.71 [0.51, 0.82]	0.55 [0.31, 0.73]
Print selectivity	Strength	0.35 [0.27, 0.71]	0.44 [0.38, 0.76]	0.28 [−0.01, 0.53]
	Lateralization	0.47 [0.21, 0.67]	0.68 [0.49,0.82]	0.33 [0.04, 0.57]

5. Discussion

The present study examined the development of N1 responses to print and non‐print stimuli in Chinese‐speaking Hong Kong children throughout their elementary school years using an accelerated longitudinal design. We investigated the relation between reading skill, measured by one‐minute word reading, and N1 specialization to Chinese characters, indexed by print sensitivity and selectivity. The first goal was to examine the age‐related changes in N1 strength and lateralization elicited by different types of stimuli, i.e., real characters, illegal characters, and visual controls, in a sample of children aged 7–11 years old. The second goal was to test the reading expertise hypothesis, i.e., improvement in reading skill leads to neural specialization to print. Specifically, we examined whether earlier reading skill can predict the subsequent level of neural specialization 2 years later and the longitudinal changes of neural specialization. The discussion is organized to follow these objectives.

5.1. The Age‐Related Changes of N1 Responses to Chinese Print and Non‐Print

The results of the GAMM analyses confirmed that the N1 to visual words declined in a linear fashion with age. The decline of character N1 also extended to illegal characters. This could reflect a generalization effect, which has been found in previous fMRI studies that showed the VWFA is also sensitive and selective to word‐like stimuli, e.g., letter strings that do not form words (Pegado et al. 2014). The current finding is also consistent with the previous Chinese research, which showed that the N1 to real and illegal characters declined as children developed (Zhao et al. 2019). The researchers further pinpointed that the decline in illegal character N1 was accompanied by an increase in behavioral efficiency, judging the lexicality of the stimuli. The results could be explained by Price and Devlin (2011)'s interactive account of the visual word neural specialization, in that the decline of N1 indicates less top‐down predictive errors and thus more neural efficiency.

Moreover, the character N1 decreased more rapidly with age than illegal character N1, which was consistently revealed across GAMM and LMM analyses in the current study. A possible explanation is that children are more familiar with real characters, and thus the child brain generates less predictive error. The faster decrease of character N1, however, resulted in a decrease in print selectivity strength with development. This finding rejects our hypothesis that print selectivity would enhance with development. The primary factor contributing to this inconsistency may be the comparison condition. Previous studies that showed development‐related enhancement in print selectivity employed comparison stimuli such as faces and limbs (Dehaene‐Lambertz et al. ²⁰¹⁸; Li et al. ²⁰¹³; Nordt et al. ²⁰²¹). Reading experience has been found to exert negative effects on the activation of these comparison stimuli in the left hemisphere (or positive effects in the right hemisphere), indicating competition effects (Dehaene et al. 2005). Therefore, the divergent reading effects between print and control conditions result in an enlarged print selectivity effect over time as reading skill improves. The present study adopted illegal characters as the comparison condition, the processing of which is influenced by reading experience in a similar way to that of real print symbols (Coch and Meade 2016; Pegado et al. ²⁰¹⁴), leading to findings that differ from previous research.

When using illegal words as comparison conditions in a novel paradigm, although Wang et al. (2024) still found an overall increase in print selectivity response over 2 years in elementary school children in alphabetic context, they further showed that children who had more reading gains exhibited a smaller increase in print selectivity, and children with better reading skills showed weaker print selectivity at T2. These results suggest a further developmental pattern that echoes our current finding, i.e., a decrease in print selectivity with development.

At last, a non‐linear relation was observed between age and the visual control N1. The control N1 decreased until the age of 10, followed by a slight increase. The earlier decrease echoed previous findings indicating a general ERP decrease with age, which could be attributed to maturational factors such as synaptic pruning and myelination in the visual area (Picton and Taylor 2007). The subsequent increase in N1 might suggest an influence other than general maturation. Older and proficient readers may begin to process strokes as linguistic units. Developmental studies have found that Chinese children cannot reliably distinguish well‐formed characters from illegal characters with one stroke missing or redundant until grade 5 (Ho et al. 2003). The top‐down processing of strokes as linguistic units, rather than as lower‐level visual information, may be reflected as an increase in N1 elicited by random combinations of strokes. That said, the sample size of older children is small, and therefore, the findings should be interpreted with caution.

Regarding N1 lateralization, the results of LMM indicated that the processing of illegal characters was more right‐lateralized than that of real characters over 2 years. This could be due to the novelty of illegal characters for children, as it has been indicated that processing of novel stimuli is right‐lateralized (Rossion et al. 2002). Alternatively, the illegal characters were created by violating the spatial relationship rules of radicals (sublexical units in Chinese characters). For instance, the illegal character “子女” can be transformed into the correct form “好” if reversing the positions of the radicals. Spatial relationships as lower‐level visual features are predominantly processed in the right hemisphere (Chung et al. 2018). The conditional difference between the real and illegal characters resulted in left‐lateralization of print selectivity across the two time points, which aligns with previous findings in Chinese elementary school children (Lo et al. 2019; Tong et al. ²⁰¹⁶). In the same vein, we speculate that the real character stimuli having phonetic radicals on the right side may contribute to the left‐lateralization found in this study. Specifically, child participants might have paid attention to the right visual fields to access pronunciation during the lexical decision task. This process might have been enhanced with improvement in reading skills.

Despite the right‐lateralization of N1 responses observed in children in earlier elementary grades, the results of GAMM and LMM consistently indicated that engagement of the right hemisphere decreased for both real and illegal characters, but not for visual controls, as children develop. This developmental pattern is consistent with previous findings in Chinese children (Tong et al. 2016) as well as in children who read alphabetic scripts (Turkeltaub et al. 2003). Altogether, the present finding falls in line with a consistent developmental pattern in visual word processing, i.e., developmental changes occur in the direction of increased left‐lateralization. This lends support to the universal account of the neural mechanism underlying reading and reading acquisition, suggesting that the left hemisphere plays a dominant role (Rueckl et al. 2015; Turkeltaub et al. ²⁰⁰³).

On the other hand, the involvement of the right hemisphere when processing Chinese characters may continue as it has been recently observed in adult readers. Yu et al. (2022) found that the right hemisphere was more active when processing real characters than pseudo characters, potentially reflecting semantic processing. Future longitudinal studies with a larger age range and multiple tasks are needed to verify this speculation.

5.2. Earlier Reading Skill Predicts Subsequent N1 Print Specialization

The current study showed that reading fluency predicted the longitudinal change in print sensitivity (strength) over the subsequent 2 years. Children with higher reading fluency at T1, as measured by the age‐and‐grade‐regressed standardized score, showed more decrease in strength over 2 years compared to less fluent readers. According to the inverted‐U‐shape model, print sensitivity in readers at the intermediate stage of reading acquisition tends to increase less or decrease in strength with development (Fraga‐González et al. ²⁰²¹; Maurer et al. ²⁰⁰⁶; Price and Devlin ²⁰¹¹). Our finding supports this model by showing that more fluent readers decreased in print sensitivity and thus experienced a more advanced development, while less fluent readers showed an increase in print sensitivity. This accounts for the nonsignificant longitudinal change at the group level, which was canceled out by individual variations. The reduced strength in N1 print sensitivity may reflect functional efficiency of the reading‐related neural network due to fewer predictive errors, as previously discussed in the framework of the interactive account of VOT's contribution to visual word recognition (Price and Devlin 2011); this functional efficiency is likely driven by reading skill (increased automaticity).

In terms of print selectivity, reading skill did not explain the individual variation of longitudinal change in strength. Our results showed that print selectivity strength had weak within‐session intraindividual stability at both time points. Therefore, print selectivity strength, as an individual measure of neural print specialization, was relatively unreliable and contained much noise, potentially masking its association with reading skill. In contrast, print sensitivity (both strength and lateralization) consistently showed good reliability across time. This falls in line with the prediction of the IS theory (Johnson 2011) that print selectivity develops later than print sensitivity. Although the former is present at the group level, it might take more reading experience for print selectivity to consolidate, so that it can serve as a measure of an inter‐individual trait as stable as print sensitivity.

Regarding lateralization, higher earlier reading skill was linked to more left‐lateralized print sensitivity and selectivity 2 years later. This finding extends previous evidence that shows a concurrent association between reading skill and the left‐lateralization of print‐sensitive and selective responses (Maurer et al. 2011; Wang et al. ²⁰²²). This longitudinal prediction provides stronger evidence than concurrent associations for a causal relation, suggesting that improvements in reading skill led to neural functional efficiency for processing print and greater left‐lateralization of the neural reading network. We presume that fluent readers have more reading experience over time, depicted as the Matthew effect in reading development (Cunningham and Stanovich 1990). They are likely to receive more explicit reading instruction given that they are faster and participate in more reading and writing activities in and outside of school. Consequently, they develop more advanced neural specialization for print.

Part of this causal relation has been demonstrated in training studies. For example, intensive training of grapheme‐phoneme correspondence rules using an artificial script increased left‐lateralization of print sensitivity (Yoncheva et al. 2015). A training study targeting children with dyslexia demonstrated that improved readers (in reading fluency) showed reduced activation in real word N1 (Fraga González et al. 2016), consistent with the current observation of reduced print sensitivity in more fluent readers. The present study tentatively suggests more reading experience (higher reading skill) leads to greater left‐lateralized print selectivity for elementary school children. Future training studies could examine the enhanced neural selectivity for print as a potential outcome.

The present study has several limitations. First, there was an issue of undersampling of children at both ends of the age distribution. Thus, the findings regarding the developmental trajectories should be interpreted with caution, and the findings of longitudinal predictions are biased toward children in grades 2 and 3. Second, although the current analysis indicated a substantial correlation between twins and co‐twins on N1 measures, the small sample size prevents us from further analyzing the relative contributions of environmental versus genetic factors to the development of N1 print specialization. Third, the real character stimuli in the EEG experiment were all compound characters. Whether the findings can be generalized to single‐component characters remains unknown. Without phonetic cues on the right side, it is possible that the left‐lateralization of character specialization could dim for single characters. Future research is needed to explore this issue.

In conclusion, the present study described the process of neural specialization for Chinese print throughout the elementary school years. Functional neural efficiency for print processing, as indicated by the reduced print‐related N1, improved over time. This improvement was accompanied by a decrease in the engagement of right hemisphere processing characters and similar stimuli. This developmental process is largely driven by improvements in reading skills, as higher reading skills at T1 predicted a longitudinal decrease in N1 print sensitivity and subsequent left‐lateralization of print sensitivity and selectivity two years later. Altogether, the current study reveals that the left hemisphere becomes increasingly prominent and efficient in neural specialization for Chinese print, supporting the linguistically universal account of the neural mechanisms for reading development.

Supporting information

Data S1.

Data Availability Statement

Data and analysis script is available upon request.