Researchers at the University of Hawaiʻi at Mānoa have made strides in utilizing AI to understand endangered languages, aiding communities in language preservation and technology accessibility. The study, led by PhD graduate Kaiying Lin and Professor Haopeng Zhang, introduces FORMOSANBENCH, a benchmark for evaluating AI systems on low-resource Austronesian languages. Focusing on Atayal, Amis, and Paiwan—languages from Taiwan facing extinction—this research tested AI capabilities in machine translation, speech recognition, and text summarization. The results indicate significant performance gaps between these endangered languages and widely spoken ones like English. Despite attempts to fine-tune AI models, they struggled to meet the needs of low-resource languages. The research aims to drive future technology developments towards inclusivity, supporting language preservation. All datasets and code have been made publicly available to encourage further research. This pivotal study will be presented at the 2025 Conference on Empirical Methods in Natural Language Processing in Suzhou, China.
Source link
Share
Read more