META in partnership with UNESCO launches a new data collection program to improve speech and translating AI

LTPP (Language Technology Partner Program – a Language Technology Partnership program) is a joint initiative of UNESCO and META✴ for the search for authors that can provide more than 10 hours of speech records with transcriptions, large volumes of written text and sets of translated texts in different languages. In the future, these data will be integrated into AI models with an open source for speech recognition and translation.

Image source: UNESCO

LTPP efforts will be focused on insufficiently served languages ​​to support the work already carried out in this direction of UNESCO. “Ultimately, our goal is to create intellectual systems that can understand and respond to the complex needs of a person, regardless of language or cultural origin,” said Meta✴ representative.

In addition to the new initiative, Meta✴ published an open source code of the program to assess the performance of language translation models. The test consisting of proposals created by Linguists supports seven languages, and is available on the Huging Face development platform.

Meta✴ continues to expand the number of languages ​​supported by its AI models and develop automatic translation functions for content creators. In September last year, the company began testing a tool for transferring votes on Instagram✴ Reels, which duplicates the creator’s speech in another language with automatic lipsink.

To date, processing on META✴ content platforms in languages ​​other than English is far from perfect. According to some reports, on the social network Facebook✴ 79 % of Covid misinformation in Italian and Spanish were not recognized and marked by the system, compared with 29 % in English. And messages in Arabic, on the contrary, are often mistakenly marked as inciting hatred.

Meta✴ said she was taking measures to improve her technologies for translation and moderation. And although the company positions both its language initiatives as philanthropic, there is no doubt that it is Meta✴ that will be the main beneficiary of these programs, which can significantly improve the quality of speech and translation recognition.