--- title: "Putting sign language AI into users’ hands" url: https://daily.dev/posts/putting-sign-language-ai-into-users-hands-dijs4bj0f source_url: https://deepmind.google/blog/putting-sign-language-ai-into-users-hands type: article source: "DeepMind" published: 2026-08-13T17:26:01.778Z updated: 2026-08-13T17:26:50.233Z tags: ["computer-vision", "accessibility", "google-deepmind"] reading_time: 8 upvotes: 0 comments: 0 language: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Putting sign language AI into users’ hands **[DeepMind](https://daily.dev/sources/dm)** · 8 min read · 0 upvotes · 0 comments ## Summary Google DeepMind and Android teams introduce SL2T, a massively multilingual sign-language-to-text model trained on over 100,000 hours of data across 50+ sign languages. It powers new ASL-to-English dictation features in Gboard and Live Transcribe, launching on Pixel 11. The model processes pose landmarks (via on-device MediaPipe Holistic) rather than raw video for privacy, translates directly to text without intermediate glosses, and achieves a zero-shot score of 70 BLEURT on the FLEURS-ASL benchmark, reportedly surpassing prior reported results. The project was developed with input from Deaf communities and an advisory committee (AISLAC), with plans to expand to more languages and devices. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is Google's SL2T model and what does it do for sign language translation? SL2T is a massively multilingual sign-language-to-text translation model from Google DeepMind, trained on over 100,000 hours of data spanning more than 50 sign languages, roughly a quarter of it in American Sign Language. It translates pose landmark sequences directly into text without using intermediate gloss annotations, and powers ASL-to-English dictation in Gboard and Live Transcribe on Pixel 11. _Developers tracking new multilingual and accessibility-focused AI models follow releases like this on daily.dev._ ### How does SL2T protect user privacy when translating sign language on a phone? SL2T processes sign language as a sequence of pose landmark coordinates rather than raw video, using an on-device model called MediaPipe Holistic to track points on the signer's body. Only these geometric coordinates are sent to the server for translation, and the original video can be discarded immediately after landmarks are extracted. _Engineers designing privacy-preserving on-device ML pipelines can follow approaches like this on daily.dev._ ### How well does SL2T perform on ASL translation benchmarks compared to prior models? SL2T achieves a zero-shot score of 70 BLEURT on the FLEURS-ASL (sd-test) benchmark for ASL-to-English translation quality, described as significantly higher than any previously reported score. It also addresses practical deployment issues such as streaming latency, hallucination on non-signing inputs, fairness for left-handed signers, and one-handed signing while holding a phone. _Anyone evaluating multilingual translation model benchmarks can track results like this on daily.dev._ --- Tags: [#computer-vision](https://daily.dev/tags/computer-vision), [#accessibility](https://daily.dev/tags/accessibility), [#google-deepmind](https://daily.dev/tags/google-deepmind) [View this post on daily.dev](https://daily.dev/posts/putting-sign-language-ai-into-users-hands-dijs4bj0f)