This study examines how orthographic knowledge modulates semantic processing across auditory and visual modalities in Chinese, a nonalphabetic script with opaque sound-spelling mappings. Forty-eight native Mandarin speakers performed semantic-relatedness judgements on disyllabic word pairs. Orthographic overlap (shared vs. non-shared character) and common syllable position (first vs. second) were systematically manipulated across auditory and visual sessions, with word frequency, word class, and other confounds carefully controlled. Results revealed a striking cross-modal dissociation. Robust orthographic facilitation emerged in auditory processing but was absent in visual processing. In audition, this facilitation was substantially stronger when the shared character occurred in the second syllable, where phonological disambiguation demands peak, than in the first syllable. Additional modality-specific patterns emerged: target word frequency selectively modulated reaction times in audition, whereas cross-class orthographic overlap unexpectedly produced response interference in vision. These findings indicate that orthographic activation is selectively recruited during Chinese spoken word recognition to resolve lexical ambiguity during temporally extended auditory input. The results support interactive activation models while underscoring that lexical access pathways are both modality-specific and script-specific.