Google Brings Sign-to-Text ASL Input to Gboard and Live Transcribe
SL2T 1.0 reaches Pixel 11 with explicit limits against high-stakes use or replacing qualified interpreters.
Google DeepMind introduced SL2T 1.0 on August 12 as an American Sign Language-to-English input system for Gboard and Live Transcribe. The first release is available on Pixel 11, with Google saying more devices and sign languages will follow.
For Deaf and hard of hearing people who use ASL, the practical change is that signing can now replace typing in ordinary phone tasks such as search, messages, notes and low-stakes conversations. The launch also creates an unusually clear test of whether a research model can become a useful accessibility feature without being mistaken for a qualified interpreter.
From a camera view to editable English text
Google says SL2T was trained on more than 100,000 hours of material spanning over 50 sign languages, with about a quarter of that material in ASL. On the device, MediaPipe converts the camera view into geometric landmarks representing the signer's face, body and hands. The raw video is then discarded, while the landmark sequence is sent to Google's servers for translation.
The result appears as text the signer can inspect and edit before sending it through Gboard. In Live Transcribe, the user also controls when translated text is displayed or voiced to another person. That design keeps the signer in the decision loop and positions SL2T as assistive input, not an autonomous interpreter.
The impact report draws a firm safety boundary
A joint impact report co-authored by Google DeepMind and representatives of the Google-convened AI Sign Language Advisory Committee limits the release to informal, low-stakes use. It excludes medical care, legal proceedings, formal education, employment decisions and high-stakes public services because the model was not designed or evaluated for those settings.
The report also says the software should not be used to replace qualified interpreters or evade accessibility obligations. That distinction matters: SL2T translates signing into English text, but it does not perform the full two-way, contextual and accountable work of a human interpreter.
Known errors remain visible in the release record
The report documents numerical distortion, text generated when no matching sign was intended, weaker handling of facial and other non-manual grammar, inconsistent regional-sign recognition and sensitivity to lighting and camera position. The current pose pipeline follows one main signer rather than a multi-person exchange.
Context is limited too. Input clips stop after 60 seconds or when hands leave the frame, and each clip is translated without retaining a multi-turn conversation history. Those constraints explain why a user must review the text and why the feature is not suitable for decisions where a mistranslation could materially harm someone.
What the evidence supports
The launch and product boundaries are confirmed by Google's announcement and the joint impact report. Internal and advisory testing described in those sources supports a consumer release for everyday tasks, but it is not independent evidence of accuracy across the full diversity of signers, devices, environments or conversational settings.
The next useful checkpoint is broader device and language availability accompanied by independent evaluation of translation errors, hallucinated text and performance across regional, demographic and motor differences. Until then, the strongest supported conclusion is narrower: Google has moved ASL-to-English input into two consumer products, with explicit limits on where it should be trusted.
Status
Confirmed. Internal confidence: medium. The release and its restrictions are documented in official material, while broad independent deployment evidence was not available in this run.
Sources
Update note: Last reviewed 2026-08-14. We will revise this post when Google expands availability or independent evaluations establish new performance evidence.
Sources
- Google DeepMind launch announcement — official
- AISLAC joint impact report for SL2T 1.0 — research
Drafted with AI assistance from source briefs; reviewed for citation completeness and label accuracy.