Google DeepMind has built an AI model that translates sign language into written text, and it is now live on real phones. The model, called SL2T, powers a new sign-to-text feature in Gboard and Live Transcribe on the Pixel 11. It starts with American Sign Language translated into English, with more languages promised later.
The idea is simple. A Deaf user can sign in front of their phone camera anywhere they would normally type: searching the web, writing a message, or asking Gemini a question. In group conversations captured through Live Transcribe, a Deaf person can sign their reply instead of typing it out on a small keyboard while everyone waits.
This is a bigger deal than it sounds. There are an estimated 70 million Deaf and hard of hearing people worldwide, and more than 200 sign languages in active use, yet only a fraction of them have any legal recognition. Voice typing and dictation have existed for hearing users for years. Nothing comparable has existed for signing, until now.
Earlier attempts at this problem mostly failed, and the reason is instructive. Sign language gloves, which tried to read hand shapes with sensors, treated sign language like English spelled out on the fingers. That misses most of the actual grammar. Sign languages carry meaning through facial expression, head tilt, and the speed and space of a movement, not hand shape alone. A glove that only watches the hand was never going to work.
Google's model instead tracks points across the signer's whole body and face using the phone's own camera, then sends only that coordinate data, not the video, to be translated. That is a sensible privacy choice, since it means the original footage never has to leave the device or sit on a server.
There is a business angle here that goes beyond accessibility goodwill. Companies that need sign language interpreters for meetings, customer service, or training typically pay somewhere between 18 and 150 dollars an hour for that service. Deaf employees separately report lifetime wage gaps stretching into the hundreds of thousands of dollars, tied in part to communication friction with hearing colleagues and managers. A free translation tool built into a mainstream phone will not close that gap by itself, but it lowers the cost of basic day-to-day communication in ways that add up over a career.
Apple is working on similar ground, previewing sign language interpreter support for FaceTime and publishing its own research on generating sign language output, not just reading it. That signals this is becoming a real competitive front among the big phone makers, not a one-off charity feature.
The honest caveats matter too. The feature only works on the Pixel 11 for now, only translates American Sign Language, and Google's own examples show it still stumbles on rare signs, fast fingerspelling, and some grammatical structures. It is a genuine first step, not a finished product. Whether it becomes useful at scale depends entirely on how fast Google can get it onto cheaper phones and into more of the world's 200-plus sign languages.