Fig 1.
Design in the behavioral and ERP experiments.
(A) Examples of stimuli with decreasing visual-orthographic regularity for adult skilled readers in Chinese script: real, pseudo, false characters, and stroke combination. These stimuli are valid at different levels of visual units (i.e., stroke, radical, and character). Stroke combinations were constructed by combining strokes of characters that do not form valid radicals (invalid at radical level). False characters consisted of valid radicals, but they were placed in illegal positions and did not form valid characters (invalid at character level). Pseudo characters consisted of radicals at orthographic legal positions, but the resulting “characters” do not exist in the lexicon (frequency = 0). Real characters were valid at all levels of visual units. The two different colors in the examples are for demonstration only to show the two radicals. (B) Schematic depiction of the lexical decision task in the behavioral experiment (left panel) and one-back color matching task in the ERP experiment (right panel). In the lexical decision task, children were instructed to judge as accurately and as fast as possible whether a stimulus was a real Chinese character or not. In the one-back color matching task, each stimulus was presented in one of 3 colors (red, green, or yellow). Children were asked to press a button as accurately and as fast as possible whenever the same color occurred twice in a row. ERP, event-related potential.
Fig 2.
Development of orthographic regularity perception and lexical classification efficiency for 4 types of stimuli.
(A) Percentage of character response. Hit for real characters and false alarm for the other 3 stimulus types in the behavioral lexical decision task. The significance tests were two sided, and they were binominal tests against chance level (50%). The percentages of character response to real and pseudo characters were high above chance level; the percentage of character response to stroke combinations was well below chance level at 3 ages. Importantly, 7-year-olds showed a higher percentage of character response to false characters than chance level. The percentage of character response to false characters was closest to chance in 9-year-olds. With more reading experience, the percentage of character response to false characters was lower than chance level. (B) Reaction time data of character responses; 7- and 9-year-olds showed no reaction time differences between pseudo and real characters, while 11-year-olds showed longer reaction times when classifying pseudo characters as real characters. Lines in the boxes denote means; **p < 0.01, *p < 0.05. See 10.6084/m9.figshare.8948912 for subject data.
Fig 3.
(A) Orthographic regularity perception and lexical classification efficiency in children across 3 stages of learning to read (also see Fig 2). Hypothesized neural N1 responses in feature detection model (B) and predictive coding model (C). (D) Neural response of the system with age decrease due to general brain maturation. (E) Hypothesized development of neural responses to stimuli with different orthographic regularities at 3 stages of learning to read based on the feature-detection model. In the beginning readers (7 years old), neural response to real/pseudo/false characters with higher orthographic features would be stronger than that to stroke combinations. In the middle stage (9 years old), the neural response to false characters would be reduced and weaker than that to real/pseudo characters with higher orthographic features. With further reading training, in the later stage (11 years old), neural response to pseudo characters (less regular) would be reduced and weaker than real characters. (F) Hypothesized development for the predictive coding model. In the beginning stage, the neural response to real/pseudo/false characters would be stronger than that to stroke combinations (fewer predictions and prediction errors). In 9-year-olds, neural response to false characters (lower efficiency of prediction due to higher prediction error) would be increased and stronger than that to real/pseudo characters. In 11-year-olds, neural response to pseudo characters (lower efficiency of prediction due to higher prediction errors) would be stronger than that to real and false characters.
Fig 4.
Development of N1 responses to stimuli with different orthographic properties at 3 stages of learning to read.
(A) Topographic maps at N1 peaks of 4 types of stimuli on P7 electrode across 3 age groups (see S8 Table and S4 Text for detailed results of N1 peak latency). (B) Grand ERP waveforms of 4 types of stimuli at occipitotemporal electrodes in each group (left panel: O1/P7/TP9 electrodes on the left hemisphere; right panel: O2/P8/TP10 electrodes on the right hemisphere). Error bands are 95% nonparametric CIs (2,000 bootstraps) (C) Box plots of mean N1 amplitudes for the 4 stimulus types at left occipitotemporal electrodes in each group. At the beginning of learning to read (7 years), no N1 difference was observed among the 3 types of stimuli (real, pseudo, and false characters) that were perceived as character-like. After about 2 years of reading training (9 years), N1 responses were observed to be greater for lower orthographic-regularity false characters that were not predicted efficiently (high prediction error) than for real and pseudo characters, whereas no difference was found between the latter two types of stimuli. With further reading training (11 years), pseudo characters with lower efficiency of prediction (higher prediction error) evoked a stronger N1 than false and real characters. Lines in the boxes denote means; **p < 0.01, *p < 0.05, †p < 0.1. See 10.6084/m9.figshare.8948912 for subject data. ERP, event-related potential.
Table 1.
Participants information and their reading performance.