Putting sign language AI into users’ hands
deepmind ·
Google DeepMind has launched SL2T, a multilingual artificial intelligence model designed to translate sign language into text for Deaf and hard-of-hearing individuals. Initially rolling out with American Sign Language to English capabilities, the feature is embedded within Gboard and Live Transcribe applications on Pixel 11 smartphones. Because sign languages operate as independent linguistic systems with unique grammars, the technology relies on advanced computer vision to decode simultaneous hand, facial, and bodily movements rather than simply mapping signs to spoken words. To safeguard privacy, the system converts physical gestures into geometric coordinate points before processing, effectively eliminating the need for raw video storage and skipping outdated intermediate translation steps. Built using more than 100,000 hours of footage across fifty distinct sign languages, the architecture achieves top-tier benchmark results while optimizing for practical concerns like low latency and left-handed usage. Development was heavily guided by Deaf community advisors and experts to ensure cultural relevance, ethical deployment, and transparent reporting of both strengths and limitations. The company intends to broaden language support and device compatibility in the future while keeping the technology freely available to advance global digital inclusion.
deepmind