Meta Unveils NLLB-200 AI Model to Translate 200 Languages
Meta introduces NLLB-200, a single AI model that translates 200 languages with significantly higher accuracy than previous technology. The company openly releases the model and dataset to help bridge the digital language divide.
Meta reveals a breakthrough in machine translation with the launch of NLLB-200, a single AI model capable of translating 200 different languages. This new technology scores an average of 44% higher in translation quality compared to previous AI systems, with accuracy improvements exceeding 70% for certain African and Indian languages.
To support and evaluate this massive undertaking, Meta builds the FLORES-200 dataset, which tests AI performance across 40,000 different language directions. The company openly shares the NLLB-200 models, the FLORES-200 dataset, and the training code with developers to encourage further innovation in the field of language technology.
This advancement directly powers more than 25 billion daily translations across Facebook, Instagram, and other Meta technologies. Additionally, Meta offers up to $200,000 in grants to nonprofits and researchers who apply NLLB-200 to initiatives focused on sustainability, education, and other UN Sustainable Development Goals.