Google Introduces SL2T AI Model for Translating Sign Language into Text
Google DeepMind has developed an innovative AI model called SL2T, which can translate sign language into text in real time. The technology is designed for the 70 million people worldwide who use sign language. The first users to try the system will be owners of Pixel 11 smartphones through the Gboard and Live Transcribe applications. At launch, the model supports American Sign Language (ASL) with translation into English.
Full Smartphone Control and Communication
With SL2T, deaf and hard-of-hearing users can interact with their smartphones without manually entering text. Gestures can now be used to:
• search for information online and write emails;
• communicate through messengers and social media;
• give commands to the Gemini AI assistant;
• translate sign language directly during video calls.
How It Works and Data Protection
Unlike similar systems, SL2T tracks not only hand movements but also facial expressions, head movements and body posture, which are critical for conveying meaning accurately. The model is adapted for left-handed and right-handed users, as well as for people who use only one hand.
The developers have also paid particular attention to privacy: users’ video recordings are not uploaded to the cloud. The MediaPipe Holistic algorithm captures only anonymous coordinates of facial and body landmarks, sending only this digital data to the server.
In the future, Google plans to add support for other sign languages around the world and teach the AI to generate video responses in sign language.
ORIENT
