Meta and UNESCO Join Forces to Improve AI Translation and Speech Recognition

Meta, the parent company of Facebook and Instagram, has teamed up with UNESCO (the United Nations Educational, Scientific and Cultural Organization) on an ambitious project to significantly enhance artificial intelligence (AI) for translation and speech recognition. This collaboration aims to bridge the language gap and empower speakers of under-resourced languages globally.

The Language Technology Partner Program

At the heart of this initiative is Meta’s newly launched “Language Technology Partner Program.” This program seeks external collaborators to contribute valuable linguistic data, crucial for training and improving AI models. The program is particularly focused on languages that haven’t benefited from extensive technological development, aligning perfectly with UNESCO’s goals.

What kind of contribution is needed?

Meta is requesting contributions of the following types:

• Speech Recordings with Transcriptions: Contributors are asked to provide at least 10 hours of audio recordings in their target language, accompanied by accurate written transcriptions. This data helps train AI models to accurately recognize and understand spoken words.
• Large Written Texts: Significant amounts of written text are needed, specifically texts containing 200 or more sentences. This provides crucial context and grammatical patterns for the AI to learn from.
• Sets of Translated Sentences: Parallel text – that is, identical sentences translated into multiple languages – is also vital. This allows the AI to learn the relationships between different languages, improving the accuracy of translation.

These contributions are vital for creating comprehensive and accurate AI translation and speech recognition models.

Focusing on Underserved Languages

A key focus of the program is on supporting “underserved languages.” These are languages with limited digital resources and technological support, often spoken by smaller communities or Indigenous populations. This aligns directly with UNESCO’s commitment to preserving and promoting linguistic diversity, a critical component of cultural heritage.

A Real-World Example: The Nunavut Initiative

The partnership between Meta and UNESCO has already yielded concrete results. The government of Nunavut, a northern Canadian territory, has joined the program. The project aims to develop robust translation systems for Inuktitut and Inuinnaqtun, two Inuit languages spoken in the region. This initiative showcases the program’s potential to empower Indigenous communities by providing technological access to their languages. The project specifically supports UNESCO’s International Decade of Indigenous Languages, highlighting the importance of preserving and promoting these vital linguistic heritages.

BOUQuET: An Open-Source Benchmark for Translation AI

To further advance the field of AI translation, Meta is releasing an open-source benchmark called BOUQuET. This benchmark is a standardized test designed to objectively evaluate the performance of different AI translation models. The sentences within BOUQuET have been carefully crafted by linguistic experts to ensure comprehensive coverage of various linguistic structures and complexities. This allows for fairer and more accurate comparisons of the various AI models currently on the market.

The BOUQuET benchmark is available for contributions and collaborative development. This open-source approach encourages wider participation and improves the overall quality of AI translation technology. This collaborative effort aims to establish a universally accepted standard for measuring the performance of AI translation systems creating a level playing field for developers and researchers alike. Anyone interested in contributing can access the platform and engage in this valuable collaborative effort.

Meta’s Broader Commitment to AI Translation

Meta’s involvement in AI translation is not merely a matter of corporate social responsibility. It aligns strategically with its core business model of connecting people globally. Accurate and efficient translation is paramount for fostering effective communications across diverse linguistic backgrounds. The company’s investment in AI translation reflects a long-term commitment to breaking down communication barriers.

Existing AI initiatives at Meta

Meta has already made significant strides in this area. An example is a tool it developed that uses AI to automatically dub Reels (short-form videos on Instagram) into other languages while maintaining lip-sync accuracy. Initially rolled out for English and Spanish videos in the US, the technology promises to make video content accessible to wider audiences. Furthermore, Meta’s AI assistant is now available in 43 countries and more than a dozen languages, indicating its global reach and commitment to multilingual accessibility.

In conclusion, the partnership between Meta and UNESCO represents a significant step forward in the development of advanced AI translation and speech recognition technologies. By focusing on underserved languages and fostering open collaboration, this initiative promises to empower communities, preserve linguistic diversity and revolutionize communication on a global scale. The creation and open-source release of the BOUQuET benchmark for assessing AI models further exemplifies the commitment to transparency and collaborative improvement within the field. This initiative offers a beacon of hope for a future where language is no longer a barrier to communication and access to information.