Meta’s new translation system can translate 200 different languages. The universal machine translator is also expected to overcome hurdles in the Metaverse future.
Machine translation has gotten much better recently thanks to breakthroughs in natural language machine processing. Companies like DeepL are competing with human translators with high-quality machine translations. Tech giants like Google and Meta are also developing their own AI systems for translation, mainly to make content more accessible on their platforms like YouTube, Facebook and Instagram. But the systems trained for AI translation need data – and that’s scarce for much of the world’s spoken content. Researchers distinguish between so-called high-resource and low-resource languages – those for which there are already very many translations on the Internet, such as English, and those for which there are almost no translations.
Meta’s “No Language Left behind” seeks the universal translator Meta CEO Mark Zuckerberg, who wants to connect as many people as possible – currently still on Facebook and Instagram, in the future in the Metaverse – therefore sees the development of a “Universal Speech Translator” as an important task for his company. In fact, Meta has been researching machine translation for years. In 2018, for example, it achieved great success with unsupervised trained AI systems and retranslation. In 2020, Meta introduced M2M-100, a system that can translate 100 languages. In 2021, an AI system based on it became the first multilingual AI model to achieve the highest score on the WMT2021 translation benchmark. Full story here