A DeepSeek-R1 fine-tuning approach for missing character inference in Dunhuang ancient Tibetan texts

Tibetan ancient books contain rich historical, philosophical, and medical knowledge, but many manuscripts are fragmented and illegible due to poor preservation. Traditional manual repair relies on expert judgment, which is inefficient and subjectively biased. This article proposes a method for inferring missing characters in Dunhuang ancient Tibetan texts by fine-tuning the DeepSeek-R1 large-scale language model. By integrating large modern Tibetan and ancient Tibetan texts from Dunhuang, a reasoning model, DeepSeek-R1-Tibet, suitable for the semantic context of ancient Tibetan, was constructed. The experimental results demonstrate that fine-tuned DeepSeek-R1-Tibet achieves the largest performance gains across all conditions, outperforming CINO, mBERT, and TisBERT on the Hits@3 metric in the three-syllable condition (0.667 vs. 0.456 for mBERT), where generative reasoning over broader context proves increasingly advantageous. This work has opened up new avenues for the digital preservation and intelligent organization of Tibetan ancient books.