NüshuRescue: Reviving the Endangered Nüshu Language with AI (aclanthology.org)

NüshuRescue is an innovative AI-driven framework aimed at preserving the endangered Nüshu language, historically utilized by Yao women in China to express themselves in a patriarchal society. Recognizing the difficulties of reviving low-resource languages, this framework enables training of large language models (LLMs) with minimal linguistic data. Central to the project is the NCGold dataset, a groundbreaking 500-sentence Nüshu-Chinese parallel corpus that's the first publicly accessible resource of its kind. By using GPT-4-Turbo, NüshuRescue achieved an impressive 48.69% translation accuracy from just 35 examples from NCGold, further producing NCSilver, which consists of 98 translated modern Chinese sentences. This effort is significant for the AI/ML community as it showcases how advanced models can be applied to linguistic revitalization, streamlining processes that traditionally require extensive human resources. Additionally, the development of FastText-based and Seq2Seq models enhances avenues for further research on Nüshu, emphasizing the framework's versatility. All datasets and code have been released publicly, positioning NüshuRescue not only as a vital tool for language preservation but also as a reference point for future projects focused on endangered languages.