Meta launches new program to improve speech and translation AI

Meta, in collaboration with UNESCO, is launching an innovative program aimed at collecting speech recordings and transcriptions. This initiative, known as the Language Technology Partner Program, seeks to gather more than 10 hours of speech recordings with transcriptions, extensive written text, and sets of translated sentences in diverse languages. The goal is to enhance AI speech recognition and translation models, which will ultimately be open-sourced for public use.

The program is particularly focused on underserved languages, aligning with UNESCO’s mission to support linguistic diversity. One of the initial partners is the government of Nunavut, a sparsely populated territory in Northern Canada, where residents speak Inuit languages collectively known as Inuktut.

Meta’s efforts are philanthropic in nature, aiming to develop intelligent systems that can understand and respond to complex human needs, regardless of language or cultural background. This initiative comes as Meta continues to expand the number of languages its AI-powered assistant supports and tests new features like automatic translation for creators.

However, Meta has faced criticism for its handling of content in languages other than English. Reports indicate that Facebook left nearly 70% of Italian- and Spanish-language COVID misinformation unflagged, compared to just 29% of similar English-language misinformation. Additionally, leaked documents reveal that Arabic-language posts are often erroneously flagged as hate speech.

In response, Meta has stated it is taking steps to improve its translation and moderation technologies. The new program with UNESCO is a significant step forward in this endeavor, focusing on the development of more inclusive and accurate AI models.