Published: 13 August 2026
Google DeepMind announced SL2T (sign-language-to-text) on 12 August 2026, a massively multilingual translation model that brings sign-to-text dictation to Gboard and Live Transcribe. The model launches first for American Sign Language (ASL) to English and reaches consumers on 20 August 2026, when the Pixel 11 ships.
What can SL2T do?
SL2T lets Deaf and hard of hearing users sign to their phone anywhere they would normally type. In Gboard, users can sign to search the web, draft messages or documents, or ask Gemini to answer a question or carry out a task. In Live Transcribe, users can sign their side of a conversation instead of typing back and forth with a hearing person.
The model is built to handle real usage rather than only lab conditions. Google DeepMind says it minimised streaming latency, worked to prevent the model from hallucinating text when the camera picks up non-signing movement, and tuned performance for the roughly 10 percent of signers who are left-handed as well as for one-handed signing, which is common while a user holds the phone in the other hand.
How does SL2T work?
Unlike spoken-language transcription, which maps sound to text in the same language, sign languages are independent languages with their own grammar and lexicon. SL2T is therefore built as a true machine translation system rather than a sign-to-word converter, and it also has to interpret physical movement: simultaneous motion of the hands, arms, torso, head and face, tracked at high frame rates.
To protect privacy, SL2T does not send video to Google’s servers. An on-device model called MediaPipe Holistic tracks pose landmark points on the signer, and only those geometric coordinates are transmitted for translation, after which the original video is discarded. SL2T translates the coordinate sequence directly into text, skipping the intermediate gloss annotations used in earlier sign language research, which Google DeepMind says fail to capture non-manual markers and spatial constructions.
SL2T benchmarks and technical specs
SL2T is trained on more than 100,000 hours of data spanning over 50 sign languages, with roughly a quarter of that data in ASL. Google DeepMind says training jointly across languages, dialects and proficiency levels helps the model learn shared underlying structure and outperform models trained on a single language.
On FLEURS-ASL (sd-test), a benchmark for ASL to English translation quality, Google DeepMind reports a zero-shot score of 70 BLEURT, which the company describes as significantly higher than any previously reported score on that benchmark. This figure is vendor-reported: Google DeepMind has not published a comparison table of prior scores or named a specific competing model it is measured against, so the claim should be read as self-reported rather than independently verified.
How does SL2T compare to earlier sign language technology?
Sign language AI has historically lagged spoken language AI by a wide margin. Google DeepMind notes that more than 200 sign languages are in use worldwide by an estimated 70 million Deaf and hard of hearing people, and that earlier approaches such as sign language gloves were fundamentally limited because sign languages are not simply spoken language mapped onto the hands. There is no widely used competing consumer product that performs the same sign-to-text function at comparable scale, so SL2T is best understood as an early entrant in a largely unaddressed category rather than a head-to-head replacement for an existing model.
SL2T availability and rollout
SL2T ships inside Gboard and Live Transcribe starting on the Pixel 11, which becomes available on 20 August 2026, at no additional cost. Google DeepMind says more devices are coming soon, along with support for additional sign languages beyond ASL.
The release was shaped by the AI Sign Language Advisory Committee (AISLAC), a group of Deaf organisations and subject-matter experts that Google DeepMind says it worked with on evaluation and impact assessment. Alongside the model, Google DeepMind and AISLAC co-authored a joint impact report detailing SL2T’s capabilities and current limitations, including occasional errors on rare signs, rapid fingerspelling, passive constructions and classifier depictions.
For more details visit the official Google Deepmind announcement on SL2T.
