Meta, the American technology company, has introduced a revolutionary artificial intelligence model named NLLB-200, capable of translating 200 different languages. This advancement is particularly significant for low-resource languages, which traditionally struggle with limited available data for effective translation.
Current AI translation models rely on artificial neural networks and require extensive training data, often sourced from the internet. However, obtaining sufficient data for some languages is both challenging and costly, leaving these languages underserved in the digital age.
Meta's innovative cross-language technology allows neural translation models to utilize their existing capabilities in high-resource languages to improve translations for low-resource ones. The creation of NLLB-200 marks a substantial leap forward in multilingual translation technology and was recently featured in the prestigious journal Nature.
Researchers highlight that NLLB-200 can significantly aid users of low-resource languages by enhancing their access to the internet and educational materials, such as books and academic papers. Despite its potential, the team acknowledges that mistranslations may still occur, emphasizing the need for continuous improvement and human oversight.
Reference(s):
cgtn.com
