To make content accessible across different languages and platforms that are owned by Meta and also to enable a technology that can support virtual experiences on metaverses in the near future The AI model NLLB 200 has been created for online communities who prefer to use their own languages on their platform.
AI researchers developed No Language Left Behind (NLLB) to create top-quality machine translation capabilities that can be used for all of the world’s languages. Meta has released an update to NLLB which will introduce an all-in-one AI model called NLLB-200 that can translate 200 languages, with more precise results than previous technologies could achieve According to Meta.
In comparing their quality in translation with prior AI research the NLLB 200 scored an average of 44% better. For a few African and Indian-based languages the translations of NLLB-200 were over 70% more precise.
To ensure that the model is evaluated and improved by NLLB 200, the company created FLORES-200, a data set that allows researchers to evaluate the AI model’s performance across the 40,000 languages that it is available. FLORES 200 allows measuring the performance of NLLB200 in each language, ensuring it is of good quality.
In order to assist other researchers to develop its translation software and expand on the work of others, Meta is opening up NLLB-200 models as well as the FLORES 200 dataset to developers, as well as models training codes and code to recreate FLORES-200, the model training data.
Meta has also offered the award of 200,000 USD of grants to support the impactful use of NLLB-200 for researchers and non-profit organizations that have initiatives that are focused on sustainability gender-based violence, food security education, gender-based violence, or other fields that contribute to The UN Sustainable Development Goals.
Non-profit organizations interested in using NLLB 200 to translate between two and more African languages as well as researchers in the field of linguistics, machine translation, as well as technology for language, have been invited to submit applications. These advances in research will help support over 25 billion translated words daily in Feed on Facebook, Instagram, and our other technology.