An investigation of phonological predictors in Chinese developmental dyslexia using a machine learning approach

Phonological awareness (PA), rapid automatized naming (RAN), and verbal short-term memory (VSTM) deficits have been identified as risk factors for developmental dyslexia (DD) of alphabetic orthographies. Yet, these three facets of phonological skills in DD of Chinese, a logographic writing system, have rarely been investigated together. The present study aimed to investigate PA, RAN, and VSTM deficits in Chinese children with DD. We compared 128 third and fourth grade Mandarin-speaking Chinese children with DD aged between 8 and 11 years (mean age = 9.4 years, 73 male) and 135 age-matched controls on 7 phonological tasks, including 2 PA tasks (phoneme deletion and onset/rime deletion), 3 RAN tasks (digit, object, and color), and 2 VSTM task (spoonerism and digit span). A combination of logistic regression analysis and a machine learning approach with SHapley Additive exPlanations (SHAP) methods was employed for data analyses. Logistic regression analysis showed that similar to the results from DD in six European languages by Landerl and Ramus et al. (2013), phoneme deletion and RAN digits played major roles in predicting Chinese DD, while digit span played a minor role. SHAP analysis revealed that the most effective predictor tasks for Chinese DD are RAN pictures, phoneme deletion, and spoonerism. The results suggest important roles of RAN and PA and a minor role of VSTM in predicting Chinese DD. Findings have important implications for the diagnosis and remediation of Chinese DD.