2026 Theses Doctoral

Chinese Character Perception Across Expertise Levels

Zhou, Zhuqian

Learning to recognize Chinese characters requires learners to acquire perceptual expertise with a large set of visually complex symbols whose structures differ markedly from alphabetic characters. Although decades of research have focused on Chinese literacy development, little is known about how the visual similarity structure of Chinese characters is represented at different proficiency levels or how instruction to promote the learning of Chinese characters might be optimized. A specific question raised and addressed here is whether targeted perceptual learning can shift novices toward expert-like representations. This dissertation addresses these questions by combining feature-based analyses of visual similarity with an experimental training intervention inspired by perceptual learning modules (PLMs).

Study 1 used an online similarity-sorting task with a factorially designed set of 16 Chinese characters to examine how novices, learners, and fluent readers perceive visual similarity of the characters. Participants’ item sorts were analyzed using expert-defined features, multiple regression on distance matrices (MRM), and extended-tree (EXTREE) modeling. The results showed that novices relied primarily on generic perceptual features, whereas fluent readers’ similarity structures were organized around higher-level relational properties such as canonical configurations and recurring components. Learners showed intermediate, less internally consistent representations.

Study 2 extended these findings through a secondary analysis of a large dataset of pairwise similarity ratings for 94 characters, utilizing shape descriptors (ShapeComp) proposed in computer vision research. A multidescriptor shape distance measure strongly predicted novices’ similarity ratings and moderately predicted learners’ ratings, but showed minimal correspondence to fluent readers’ judgments. This dissociation is consistent with the conclusion that novices’ similarity spaces reflect general-purpose low-level shape processing, but experts draw on abstract, linguistic-specific visual dimensions not captured by standard computer-vision models.

Study 3 tested whether targeted training can induce expert-like perceptual changes in novices. Participants were randomly assigned to one of five instructional conditions: configuration training, component training, flashcard study, direct instruction, or a no-training control. Pre-/post-test accuracy and fine-grained error patterns were compared across conditions and to those of fluent readers. Configuration-focused perceptual training produced the largest improvements and selectively reduced same-configuration false recognitions, while component training, flashcard study, and direct instruction reduced same-component errors. These patterns are consistent with Perceptual Learning Module (PLM) principles: high-density, feedback-driven exposure facilitated learners’ attention to diagnostic structural features and shifted recognition behavior toward expert-like patterns.

Across studies, the findings provide converging evidence that expertise in Chinese character recognition depends on a representational shift from low-level shape-based to configuration-based processing, and that PLM-style interventions can accelerate this shift in novices. The dissertation contributes to theories of category learning, visual expertise, and orthographic representation, and offers concrete implications for designing PLM-inspired instructional interventions to support Chinese character learning.