The Language Technology Partner Program (LTPP) is a joint initiative by UNESCO and Meta aimed at gathering linguistic data for AI development. The program seeks authors who can contribute over 10 hours of speech recordings with transcriptions, extensive written texts, and translated text sets in various languages. This data will be integrated into open-source AI models to enhance speech recognition and translation capabilities.
Supporting Underserved Languages
LTPP primarily focuses on underserved languages, aligning with UNESCO’s ongoing efforts in this area. A Meta spokesperson emphasized, “Ultimately, our goal is to create intelligent systems that can understand and respond to complex human needs, regardless of language or cultural background.” In addition to this initiative, Meta has released open-source code for a benchmark designed to assess language translation models. The benchmark, developed by linguists, supports seven languages and is available on the Hugging Face AI development platform.
Meta is also expanding its AI-supported languages and refining automatic translation features for content creators. In September, the company began testing a voice translation tool for Instagram Reels, enabling automatic dubbing with lip-syncing.
Challenges in Language Moderation
Despite Meta’s advancements, challenges remain in handling content across different languages, notes NIX Solutions. Reports indicate that on Facebook, 79% of COVID-19 misinformation in Italian and Spanish went unflagged, compared to 29% in English. Meanwhile, messages in Arabic were often mistakenly labeled as hate speech.
Meta has stated that it is working to improve its translation and moderation technologies. While the company presents its language initiatives as philanthropic, the primary beneficiary remains Meta itself, as these advancements will significantly enhance its AI-driven speech recognition and translation. Yet, we’ll keep you updated as more integrations become available.