AI News

NüshuRescue is an AI-driven framework for supporting the preservation and revitalization of Nüshu, a rare script historically used by Yao women in China. The framework is designed to train large language models on endangered languages with minimal data, while automating evaluation and expanding target corpora. Its foundation is NCGold, a 500-sentence Nüshu-Chinese parallel corpus described as the first publicly available dataset of its kind. In an evaluation, GPT-4-Turbo had no prior exposure to Nüshu and was given only 35 short examples from NCGold; it achieved 48.69% translation accuracy on 50 withheld sentences. The system then generated NCSilver, containing 98 newly translated modern Chinese sentences of varying lengths. The authors also developed FastText-based and Seq2Seq models to support further Nüshu research. The datasets and code are publicly available through the project’s GitHub repository, enabling additional research on low-resource language technology and revitalization.