HomeAIGoogle DeepMind Launches SL2T Sign Lan
AI

Google DeepMind Launches SL2T Sign Language AI Model

Google DeepMind has introduced a sign-language-to-text model powering dictation features on the Pixel 11 for American Sign Language users.

WHAT YOU NEED TO KNOW
  • Google DeepMind released SL2T, an AI model translating sign language directly into text.
  • The model powers free ASL-to-English dictation in Gboard and Live Transcribe on Pixel 11.
  • SL2T processes privacy-preserving pose landmarks tracked on-device via MediaPipe Holistic.
  • The architecture achieved a zero-shot score of 70 BLEURT on the FLEURS-ASL benchmark.

Google DeepMind announced a multilingual sign-language-to-text translation model called SL2T on Aug. 12, 2026. The technology powers new sign-to-text dictation features in Gboard and Live Transcribe on the Pixel 11 smartphone at no additional cost, beginning with American Sign Language to English translation.

According to Google DeepMind, the system allows Deaf and hard of hearing users to sign to their phone in places where they would normally type. Users can sign to search the web, draft documents and messages, send queries to Gemini, and respond during conversations in Live Transcribe without typing back and forth.

Landmark tracking and training data

The SL2T model was trained on more than 100,000 hours of data spanning over 50 sign languages, with roughly 25% of that dataset in American Sign Language. Google DeepMind noted that joint training across diverse languages, dialects, and proficiency levels allowed the model to learn shared underlying structures, outperforming single-language models in experiments.

To preserve user privacy, the system does not send raw camera footage to external servers. An on-device tool called MediaPipe Holistic tracks physical pose landmarks across the signer's hands, arms, torso, head, and face. Only these geometric coordinate points are transmitted to servers for translation, and the original video feed is discarded immediately on the device.

SL2T translates the coordinate sequence directly into written text. This bypasses intermediate text annotations known as glosses, which fail to capture non-manual markers and spatial constructions inherent to sign languages.

Performance and governance

On the FLEURS-ASL benchmark evaluating American Sign Language to English translation, SL2T recorded a zero-shot score of 70 BLEURT. Google DeepMind designed the system to account for practical deployment factors, such as minimizing streaming latency, preventing hallucinations during non-signing periods, supporting left-handed signers, and processing one-handed signing when a user holds a phone.

Development of the model was conceptualized by Sam Sepah, a Deaf employee at Google. To guide deployment, Google DeepMind established the AI Sign Language Advisory Committee alongside global Deaf organizations and subject-matter experts.

The core SL2T model was created by a joint team from Google DeepMind and Android, including Garrett Tanzer, Benoit Brard, Elizabeth Clark, Tim Dozat, and Chris Dyer, alongside Android integration leads including Ausmus Chang, Dayle Chiu, and Angana Ghosh. Google DeepMind plans to extend the technology to additional sign languages, sign generation, and further devices in future updates.

Xentir Media
Xentir Media NewsroomSource-backed AI and technology coverage, drafted by Xentir's automated editorial system under fixed human-set rules. See our editorial policy and AI usage policy.
J
Jomon · Founder & EditorFounder and editor of Xentir Media. Sets the editorial rules the newsroom system runs under, and is accountable for its corrections. About Jomon · [email protected]
The Xentir Brief
The developments worth knowing — one useful email.
Get the Brief →