Meta and UNESCO collaborate on enhancing translation AI

In a groundbreaking collaboration, Meta and UNESCO have joined forces to tackle one of AI’s most pressing challenges: bridging the gap for underserved languages.

This partnership, announced this week, aims to democratize access to advanced translation and speech recognition technologies, with a focus on preserving linguistic diversity and empowering indigenous communities.

Here’s what you need to know.

Meta and UNESCO’s initiative, the Language Technology Partner Program, targets languages often overlooked by mainstream AI systems.

By collecting speech recordings, written texts, and translated sentences, the program seeks to build robust datasets for languages like Inuktitut and Inuinnaqtun, spoken by Inuit communities in Canada’s Nunavut territory.

• Why it matters: Over 40% of the world’s 7,000 languages are endangered. This effort aligns with UNESCO’s International Decade of Indigenous Languages, a global push to protect and revitalize these cultural cornerstones.
• The ask: Organizations and communities are invited to contribute at least 10 hours of transcribed speech, 200+ sentence texts, and bilingual translations. In return, partners gain access to Meta’s workshops on leveraging open-source AI tools.

The Technology: BOUQuET and Open-Source innovation

Central to this initiative is BOUQuET, a new open-source benchmark designed to evaluate translation AI’s performance. Unlike traditional datasets, BOUQuET emphasizes:

• Linguistic diversity: Sentences are handcrafted by experts across 23 pivot languages, avoiding English-centric biases.
• Real-world complexity: Paragraphs of varying lengths replace isolated sentences, better reflecting how humans communicate.
• Community-driven growth: A public repository allows speakers of any language to contribute translations, even without fluency in English.

Meta plans to release BOUQuET on Hugging Face, inviting developers to test and refine their models.

This isn’t Meta’s first foray into multilingual AI. Last year, the company debuted tools to auto-dub Reels with lip-synced translations, starting with English and Spanish. Its AI assistant now operates in 43 countries and supports over a dozen languages.

By open-sourcing BOUQuET and collaborating with UNESCO, Meta positions itself at the forefront of ethical AI development balancing innovation with cultural preservation.

Challenges and Opportunities

While promising, the initiative faces hurdles:

• Data scarcity: Many indigenous languages lack digital footprints, making dataset creation labor-intensive.
• Cultural sensitivity: Machine translations risk erasing dialects’ nuances. Meta’s solution? Partner directly with communities like Nunavut to ensure authenticity.

Yet the potential is immense. Improved translation tools could empower education, healthcare, and legal systems in regions where minority languages dominate.

Meta encourages linguists, organizations, and even casual speakers to contribute through:

• BOUQuET’s platform: Add translations in any language.
• The Partner Program: Submit speech/text data for underserved languages.

Conclusion

Meta and UNESCO’s collaboration marks a pivotal shift in AI development one where technology serves humanity’s cultural richness rather than homogenizing it.

By prioritizing marginalized languages, they’re not just building better algorithms; they’re safeguarding identities. As this initiative grows, its success will hinge on global participation. After all, language isn’t just code. It’s the heartbeat of who we are.