Meta's AI research team is partnering with UNESCO to make artificial intelligence smarter. The goal is to make AI accessible to more people worldwide by improving its ability to understand and translate languages that are most often overlooked. According to the official announcement, this partnership between Meta and UNESCO will focus on "underserved languages."
Meta and UNESCO will develop new AI models under the "Language Technology Partner Program," where the company is looking for partners who can provide more than 10 hours of speech recordings with transcriptions, written texts and translated sentences in different languages. In return, partners will gain access to workshops where they can learn how to use Meta's open-source language tools. The Government of Nunavut, Canada, is one of the first participants, sharing data in the Inuit language; Inuktitut and Inuinnaqtun.
Additionally, Meta will be releasing an open-source translation benchmark called "BOUQuet," that will test how well AI models perform in translating the different languages. This benchmark tool also includes carefully crafted sentences to ensure richness of the human language. In the blog post, Meta has also encouraged people to contribute translations to improve AI's multilingual capabilities.
Meta said:
Ultimately, our goal is to create intelligent systems that can understand and respond to complex human needs, regardless of language or cultural background. As we continue in this direction, we’re excited to collaboratively enhance and expand machine translation and other language technologies.
This new initiative by Meta is built on its previous projects like "No Language Left Behind (NLLB)," which introduced a powerful translation model for many languages. More recently, Meta launched the "Massively Multilingual Speech (MMS)" project, which allows AI to transcribe speech in over 1,100 languages. Meta claims that with improvements in this technology, including the "addition of zero-shot speech recognition in 2024," it can transcribe audio in languages it has never encountered before.
Source: Meta
0 Comments - Add comment