Meta has developed a new AI model capable of translating 200 different languages. Unlike current models that rely heavily on large amounts of training data, which is scarce for low-resource languages, this model uses a new cross-language technology. This allows the AI to leverage its ability to translate high-resource languages to improve translations for low-resource languages. The model, named NLLB-200, was detailed in a study published in Nature.
Researchers highlight that NLLB-200 can help users of low-resource languages access the internet and educational materials more easily. However, they caution that the model may still produce mistranslations.
Source – CGTN