Meta said it will work with UNESCO to launch a new program, collecting speech recordings and transcriptions, and helping to develop openly available AI in the future.
The Language Technology Partner program, the Program, is looking for collaborators who can contribute over 10 hours of audio recordings, including transcriptions, large amounts of written text, and a set of translated sentences, in “various languages.” According to Meta, the partners will work with the company’s AI team to integrate these languages into AI’s speech recognition and translation models.
So far, partners include the Nunavut government, a densely populated territory in northern Canada. Some of the inhabitants of Nunavut speak the Intuit language known as Inuktut.
“Our efforts focus on underserved languages to support the work of UNESCO,” Meta wrote in a blog post provided to TechCrunch. “Ultimately, our goal is to create intelligent systems that can understand and respond to complex human needs, regardless of language or cultural background.”
Meta, which complements the new program, said it is releasing an open source machine translation benchmark to evaluate the performance of its language translation model. The benchmark, composed of sentences written by linguists, supports seven languages and allows you to embrace your face from the AI development platform.
Meta frams both initiatives as philanthropy. However, the company is in a position to benefit from upgraded speech recognition and translation models.
Meta continues to expand its number of pilot features, including AI-powered assistant Meta AI, support, and automatic translation for creators. Last September, Meta began testing a voice translation tool on Instagram reels, and announced that it would allow creators to blow speeches and auto-lip-sync.
The treatment of content in languages other than English across the Meta platform has been the target of many criticism. According to one report, Facebook left almost 70% of the misinformation of Italian and Spanish communities, compared to just 29% of similar English misinformation. And the leaked documents from the company reveal that Arabic posts are regularly mis-flagged as hate speech.
Meta says it is taking steps to improve translation and moderation technology.
