Resolving inherited ambiguity in Chinese historical phonology with Song dynasty corpus evidence

Large digitized corpora and large language models (LLMs) are reshaping how traditional philological evidence can be used in Chinese historical phonology. This study addresses a specific bottleneck: how to systematically resolve inherited ambiguity caused by polyphonic fanqie characters in Qieyun-system rhyme dictionaries. Focusing on Song-dynasty Guǎngyùn and contemporaneous genre-diverse texts, we propose Qieyun System Polyphonic-character Disambiguation (QYSPD), an LLM-assisted pipeline that combines dictionary-internal structure with variant filtering, corpus retrieval, and contextual definition inference. QYSPD infers role-valid primary fanqies and converts branching fanqie ambiguity into explicit primary paths. Posterior dominance analysis shows that most inferred primaries are strongly supported across genres, while genre-conditioned variation is structured by specialized lexical domains. Additional LLM comparison and sensitivity analysis show high primary-assignment agreement and stable JS-distance distributions, supporting the robustness of inference. QYSPD provides a reproducible framework for future work on phonological reconstruction and the diachronic modeling of sound change across periods and sources.