Among the biggest issues with current artificial intelligence (AI) systems is limited language support. The best example is ChatGPT which does not support Bahasa Malaysia and when told to use this language it starts speaking in Bahasa Indonesia which is very different. Meta has announced the No Language Left Behind 200 (NLLB-200) AI model which they say is the first language learning model (LLM) that can translate 200 of the world's major languages.
In tests performed against other LLMs, NLLB-200 performed 44% better in producing high-quality translations. NLLB-200 is currently used to help perform 25 billion translations on Facebook, Instagram and other Meta-owned platforms every day. Meta says accurate translations in more languages can help detect harmful content and misinformation.
In addition to Meta, NLLB-2 is now also used to produce content in languages that currently do not have many articles on Wikipedia. For example, the Lingala language has only 3260 articles while it is used by 45 million people in the Democratic Republic of Congo. At the same time there are over 2.5 million articles in Swedish even though it is only used by 10 million inhabitants.
Meta now gives NLLB-200 access to researchers and they provide grants of up to $200,000 in grants to NGOs to apply this LLM in the real world.