Pixel 11 is getting an input path that does not rely on speech. Users can sign in American Sign Language, and Gboard or Live Transcribe turns that signing into English text.
Google DeepMind introduced SL2T on August 12. The model first supports ASL to English on Pixel 11, with more devices and languages planned. The concrete change is that signing can now be used where a user would normally type: web search, messages, documents, Gemini prompts, and Live Transcribe replies.
Sign language is not English on hands
The core technical point is that sign languages are independent natural languages, with their own grammar, lexicon, facial cues and spatial structure. SL2T is therefore built as a translation model rather than a frame-by-frame gesture recognizer.
DeepMind reports more than 100,000 hours of training data across over 50 sign languages, with roughly a quarter in ASL. On the FLEURS-ASL ASL-to-English evaluation, it reports a zero-shot score of 70 BLEURT. The launch is also aimed at a real population: about 70 million Deaf and hard-of-hearing people use sign languages globally.
Privacy starts with landmarks
SL2T does not send the raw camera feed directly for translation. An on-device MediaPipe Holistic model tracks pose landmarks, then the coordinate sequence is translated on the server. That reduces the data surface compared with uploading full video, though pose data still deserves careful handling.
The practical tests now move beyond the benchmark. Google says the team worked on streaming latency, hallucination on non-signing inputs, left-handed signing and one-handed signing while holding a phone. Those issues decide whether this works in Gboard, not only in a lab demo.
What to watch next
The first release is narrow: ASL to English on Pixel 11. The next signal is whether Google can extend the system to more sign languages, especially lower-resource ones. The second signal is whether sign input becomes a general accessibility layer across calls, education, customer service and AI workflows.
For global readers, SL2T is not just another phone AI feature. It ties model scaling, accessibility, privacy architecture and minority-language data into one product test.
Sources: Google DeepMind official release, FLEURS-ASL benchmark materials, SiliconANGLE, CocoLoop, Engadget; checked SL2T launch scope, Pixel 11/Gboard/Live Transcribe integration, 100,000-plus training hours, 50-plus sign languages, 70 BLEURT benchmark scope, MediaPipe Holistic landmark flow, and initial ASL-to-English coverage.