Statistical learning : a new perspective on Chinese orthographic learning in normal and dyslexic readers

Abstract

Substantial theoretical and empirical research suggests that statistical learning, referring to the ability to use statistical properties in discovering the hidden structural and distributional patterns of the exposed input, plays a key role in oral language development. In contrast, the role of statistical learning in written language development, especially orthographic learning, has been less explored. Recently, there has been increased interest in exploring whether statistical learning is a potential source of dyslexic difficulties in reading. In addressing these issues, the present study compares both general statistical learning ability (as assessed with the Serial Reaction Time task, i.e., SRT task) and Chinese-specific statistical learning ability in Chinese children with developmental dyslexia, typically developing age-matched control and reading-level matched control children. Three groups of children (29 Chinese children with dyslexia in grades 4-5, 29 age-matched controls in grades 4-5, and 30 reading-level matched controls in grades 2-3) were tested on the SRT task (Experiment 1), and three Chinese-specific statistical learning tasks using artificial orthography, i.e., learning positional regularities (Experiment 2), learning sound-to-print mappings (phonetic regularities) and orthography-to-semantics mappings (semantic regularities) (Experiment 3). In the general statistical learning of SRT task, children with dyslexia showed a significant learning effect in the condition with long-exposure of repeated sequences, but not in the condition with short-exposure, while both age-matched and reading-level controls showed a significant learning effect in both short- and long-exposure conditions. In the Chinese-specific statistical learning tasks, all three groups of participants exhibited a significant learning effect in learning the positional regularities, sound-to-print mappings (i.e., phonetic radicals), and semantics-to-orthography mappings (i.e., semantic radicals). Children with dyslexia showed comparable learning performance with both age-matched control and reading-level matched children on learning the positional and phonetic regularities, but their performance on learning semantic regularities was significantly worse than the age-matched controls. Our findings suggest that statistical learning is plausibly the mechanism of orthographic learning and children with developmental dyslexia have challenges in some but not all statistical learning tasks. Our current findings shed new light on a long-lasting debate as to whether children with dyslexia have intact or impact general statistical learning ability by showing that the amount of exposure is a critical factor determining their statistical learning performance. Furthermore, our study is the first to show that the impaired statistical learning in children with dyslexia is partly due to their slow learning rate. Finally, Chinese children with developmental dyslexia appeared to have intact performance in learning phonetic regularities but they exhibited specific impairment in learning semantic regularities. Together, these findings support the newly proposed statistical learning model for Chinese orthographic learning emphasizing that statistical learning is a hidden mechanism of learning characters’ configurations, sound-to-print mappings, and semantics-to-orthography mappings in Chinese.