Handwritten Dictionary Retrieval for Digital Access to Yi Script Heritage

Abstract Traditional Yi script preserves linguistic knowledge and cultural memory, yet dictionary access can be difficult when users can visually recognize or reproduce a glyph but do not know its pronunciation, encoding, or index. We present a handwriting-based retrieval framework for improving access to Yi script lexicographic resources. The system combines SigLIP-B/16 with a lightweight residual adapter to retrieve among 2572 dictionary-linked glyph identities. In a one-shot glyph-disjoint test of 350 held-out identities and 417 handwritten queries, adaptation increased Recall@30 from 64.7% to 88.7% and Recall@10 from 53.0% to 75.1%, while reducing median rank from eight to three. The resulting prototype provides a handwriting-based entry point to difficult-to-index lexicographic resources while retaining the original dictionary and human interpretation as the basis for further consultation.