Meta together with UNESCO is launching a new program to collect language recordings and transcriptions. This will help develop future AI models accessible to everyone. The program, called the Language Technology Partner Program, is seeking partners who can provide more than ten hours of language recordings with transcriptions, large volumes of written text, and sets of translated sentences in various languages. Partners will work with Meta teams to integrate these languages into speech recognition and translation models, which will later become open to everyone.
Among the first partners is the government of Nunavut in Northern Canada, where some residents speak Inuit languages known as Inuktut. Meta notes that their efforts are aimed at supporting languages that are usually overlooked, which aligns with UNESCO’s work. The goal of the program is to create intelligent systems that can understand and respond to complex human needs, regardless of language or cultural context.
As part of the program, Meta is also releasing an open machine translation benchmark to evaluate the performance of language translation models. This benchmark, composed of linguist-created sentences, supports seven languages and is available on the Hugging Face platform for AI development. Meta positions both initiatives as philanthropic, although the company also benefits from improved speech recognition and translation models.
Meta continues to expand the number of languages supported by its Meta AI assistant and to test automatic translation features for content creators. Last September, Meta announced the start of testing a tool for voice translation in Instagram Reels, allowing creators to dub their own voice and automatically sync lips. Meta is also working on improving translation and content moderation technologies, especially after criticism regarding its handling of non-English content on its platforms.
