Google DeepMind SL2T sign language model

Google DeepMind's SL2T model is set to launch on August 20 on Pixel 11, enabling ASL-to-English sign dictation directly in Gboard and Live Transcribe. The system uses a two-stage on-device architecture: MediaPipe Holistic converts the camera feed into 130 pose landmark coordinates and immediately discards the raw video, then SL2T translates those landmarks to text. Google DeepMind says skipping intermediate gloss annotation allows translation quality to scale with data and lets the model capture non-manual markers and spatial grammar. The model was trained on over 100,000 hours across more than 50 sign languages and scores 70 BLEURT zero-shot on FLEURS-ASL; Google DeepMind calls it "the most capable sign language translation model to date according to key benchmarks like FLEURS-ASL (sd-test), which assesses ASL to English translation quality." Google established the AI Sign Language Advisory Committee (AISLAC) with representatives from the National Association of the Deaf, RIT/NTID, DPAN, and the World Federation of the Deaf; two facilitators from Google's Accessibility organization manage AISLAC's external engagements to keep advisory oversight separate from engineering. AISLAC members tested the model on pre-release Pixel devices, and Google DeepMind and AISLAC co-authored a joint impact report documenting current limitations including errors on rare signs, rapid fingerspelling, passive constructions, and classifier depictions. The feature is available at no additional cost. Expansion to more sign languages, devices, and sign language generation output is described as coming with no committed date.