Google DeepMind launched SL2T (sign-language-to-text), an AI model that allows users to search the web, write messages and navigate devices using sign language.
The technology is aimed at improving digital access for the estimated 70 million Deaf and hard of hearing people globally who use sign languages. It is debuting on Pixel 11 devices through Gboard and Live Transcribe, initially translating American Sign Language (ASL) into written English.
Sign language translation is different from voice transcription because sign languages have their own grammar, syntax and facial expressions. Meaning can be conveyed through movements of the hands, face and body, making direct translation more complex.
Google DeepMind's system uses on-device tracking to identify key geometric points across the user's hands, face and torso. It processes these numerical coordinates for translation rather than storing or transmitting raw video. The original video feed is deleted immediately.
SL2T then converts the movement coordinates directly into fluent text. The system does not rely on intermediate annotations known as "glosses", allowing it to account for facial cues and spatial movements that can carry meaning in sign languages.
Google DeepMind said SL2T was trained on more than 100,000 hours of data covering more than 50 sign languages, with roughly a quarter of the data dedicated to American Sign Language.
This model is meant for regular usage and not under experimental conditions. One can sign even when holding the mobile phone in their other hand. Google DeepMind has also ensured that it works equally well for left-handed signers.
This project was conceived by Deaf researchers and advocacy organizations. There was an internal advisory committee as well that helped to ensure that the technology met the needs of its target users.
Also Read: Google DeepMind CEO Demis Hassabis Calls for Urgent AGI Safety Rules
With SL2T, users can sign where they would normally type. This includes web searches, messages and documents, while Live Transcribe allows users to sign responses during conversations instead of typing.
Google DeepMind plans to bring the technology to more smartphone models and add support for further international sign languages in future updates. The Pixel 11 launch marks the first integration of a sign-to-text AI model into everyday consumer software at scale, according to Google DeepMind.