Google DeepMind has launched SL2T, a multilingual sign-language-to-text model that debuts on the Pixel 11 through Gboard and Live Transcribe, with ASL-to-English as the first supported language pair1,2.
The model lets deaf and hard-of-hearing users sign to their phone anywhere they would normally type — including searching, writing messages, or interacting with Gemini. Google DeepMind said it wants to bring AI to an estimated 70 million people who use sign language worldwide.
SL2T's architecture splits processing between device and server. An on-device model tracks pose points across the face, hands, arms, and torso; only those coordinates are sent to the server, meaning raw video never leaves the phone. The model handles one-handed and left-handed signing and includes what the description calls hallucination prevention, designed to avoid misreading incidental hand movements such as adjusting a grip.
The training dataset comprises more than 100,000 hours of data across 50-plus languages, with approximately a quarter of that data consisting of ASL — which is cited as the reason ASL-to-English is the only pair available at launch. Additional languages and devices are planned, though no timeline has been specified in the available sources.
ANALYSIS The privacy-oriented design — extracting skeletal pose coordinates on-device and transmitting only those lightweight signals — addresses a core sensitivity for any vision-based input system, particularly one aimed at accessibility users who may be signing in private or medical contexts. Launching inside Gboard and Live Transcribe, rather than as a standalone app, embeds sign-language input at the OS keyboard layer, which means it could function as a system-wide text-entry method rather than a siloed tool. The Pixel 11 exclusivity at launch constrains initial reach but follows a familiar pattern of Google debuting AI features on its own hardware before broader rollout.