Can we use NLP to extract information about long-dead languages from secondary sources more than a millennium old? This presentation explores the curation of a dataset based on the Jingdian Shiwen (c. 583), a monumental commentary including over 50,000 phonological annotations on ancient Chinese texts.