Details
- Google DeepMind introduced SL2T, a sign-language-to-text model that powers new sign-to-text features in Gboard and Live Transcribe on Pixel 11, initially for American Sign Language to English.
- The rollout lets Deaf and hard of hearing users sign to their phones for tasks such as search, messaging, document drafting, and Gemini prompts, with more devices and languages planned.
- SL2T was trained on more than 100,000 hours of data across over 50 sign languages, with about a quarter of the data in ASL, and it translates pose landmarks rather than raw video.
- Google says the system skips gloss-based intermediates, is designed for lower latency and fewer hallucinations, and supports real-world use cases like one-handed signing and left-handed signing.
- The company says SL2T reached a zero-shot 70 BLEURT score on FLEURS-ASL and was developed with Deaf community input, including a new advisory committee and user studies.
Impact
This is a notable accessibility milestone because it moves sign-language AI from research into consumer Android products for the first time. If Google expands beyond ASL, it could accelerate investment in multilingual sign-language datasets, on-device pose tracking, and user-centered accessibility tooling across the industry.
