Overview
- Google announced SL2T on August 12 and built the new ASL-to-English sign‑to‑text feature into Gboard and Live Transcribe on the Pixel 11 family.
- The system maps 130 face, hand, and body points on-device, then translates those landmark coordinates directly to text without using intermediate “glosses.”
- Google says raw camera footage is discarded on the phone and only the landmark coordinates are sent for translation, and users must opt in to share clips for research.
- DeepMind and its advisory committee warn SL2T has clear limits: it can fail on regional signs, fast fingerspelling, facial grammar, low light, partial framing, and it has not been fully tested on children or many motor disabilities.
- Google frames SL2T as an assistive input for everyday, low-stakes use that can let people who sign draft searches, messages, or replies, while cautioning it should not replace qualified interpreters or meet legal accommodation requirements and that broader device and language support is planned.