Meta is initiating a collaborative program with UNESCO aimed at gathering speech recordings and transcriptions, which the company asserts will facilitate the advancement of future AI technologies that are openly accessible.
This initiative, known as the Language Technology Partner Program, is on the lookout for partners willing to provide over 10 hours of recorded speech along with corresponding transcriptions, extensive written texts, and sets of translated sentences in a variety of “diverse languages.” As stated by Meta, these collaborators will engage with the company’s AI teams to incorporate these languages into AI-driven speech recognition and translation models, which will ultimately be released as open-source upon completion.
Among current partners is the Nunavut government, representing a remote territory in Northern Canada, where some inhabitants communicate using the Inuit languages that fall under the umbrella of Inuktut.
“Our initiatives particularly target underserved languages, supporting UNESCO’s mission,” Meta highlighted in a blog entry shared with TechCrunch. “Our ultimate aim is to develop intelligent systems capable of understanding and addressing complex human requirements, independent of language or cultural background.”
In addition to the new initiative, Meta announced the release of an open-source benchmark for machine translation to assess language translation models’ effectiveness. This benchmark, featuring sentences formulated by linguists, includes support for seven languages and is available for access and contributions on the AI development platform Hugging Face.
Meta presents both of these efforts as philanthropic undertakings, yet the company is poised to gain from enhanced speech recognition and translation capabilities.
Meta continues to enhance the breadth of languages supported by its AI assistant, Meta AI, while also piloting features like automatic translation for content creators. Last September, they disclosed plans to trial a tool that translates voices in Instagram Reels, enabling creators to dub their speech and automatically sync lip movements.
Criticism has been levied against Meta over its treatment of non-English content on its platforms. One report indicated that nearly 70% of COVID misinformation in Italian and Spanish went unflagged, compared to just 29% of similar misinformation in English. Additionally, leaked documents have shown that Arabic-language content often suffers from erroneous hate speech flags.
In response, Meta has committed to enhancing its translation and moderation technologies.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
