{"id":138908,"date":"2026-08-13T16:16:09","date_gmt":"2026-08-13T16:16:09","guid":{"rendered":"https:\/\/www.europesays.com\/ai\/138908\/"},"modified":"2026-08-13T16:16:09","modified_gmt":"2026-08-13T16:16:09","slug":"sign-language-ai-translation-googles-breakthrough-with-sl2t-model","status":"publish","type":"post","link":"https:\/\/www.europesays.com\/ai\/138908\/","title":{"rendered":"Sign Language AI Translation: Google’s Breakthrough with SL2T Model"},"content":{"rendered":"

Google DeepMind has rolled out a new form of sign language AI translation that lets Deaf and hard of hearing users sign directly to their phone instead of typing, in what the company is calling a genuine breakthrough for accessibility technology. The model, unveiled on August 12, 2026, is called SL2T \u2014 short for sign-language-to-text<\/a> \u2014 and it marks the first time this kind of technology has moved out of research labs and into everyday consumer apps like Gboard and Live Transcribe<\/a>.<\/p>\n

Key takeaways<\/p>\n

Google DeepMind launched SL2T, a multilingual sign language model that converts signing directly into written text.
\nIt currently supports American Sign Language (ASL) to English inside Gboard and Live Transcribe on the Pixel 11, at no extra cost.
\nThe model was trained on more than 100,000 hours of data across over 50 sign languages and scores 70 BLEURT on the FLEURS-ASL benchmark, well above previous models.
\nSL2T tracks pose landmarks instead of raw video to protect user privacy, discarding camera footage immediately.
\nDeaf community members shaped the project from the start, and Google created the AI Sign Language Advisory Committee (AISLAC) to guide its rollout.<\/p>\n

Google DeepMind launches SL2T, a multilingual sign language AI translation model<\/p>\n

SL2T is DeepMind\u2019s answer to a gap that has existed in AI for years: while spoken-language tools like dictation and automatic translation became mainstream over the past two decades, the world\u2019s more than 200 sign languages \u2014 used by an estimated 70 million Deaf and hard of hearing people \u2014 were largely left out. DeepMind describes SL2T as a \u201cmassively multilingual\u201d system built to close that gap with a meaningful jump in both translation quality and language coverage.<\/p>\n

The practical result is a new kind of dictation. Just as hearing users can speak instead of typing, SL2T lets Deaf users sign to their phone anywhere they would normally type \u2014 searching the web, drafting messages or documents, or asking Gemini to complete a task. Inside Live Transcribe, users can sign their side of a conversation instead of typing back and forth with a hearing person. According to Google\u2019s own testers, signing in ASL felt faster and more natural than typing out the equivalent in English.<\/p>\n

ASL support in Gboard and Live Transcribe on Pixel 11<\/p>\n

For now, the technology supports one language pair: American Sign Language translated into English. It\u2019s available first inside Gboard and Live Transcribe on the Pixel 11, with Google promising that more devices and additional sign languages are on the way. Notably, the feature is offered free of additional cost to users, positioning it as a standard accessibility function rather than a premium add-on.<\/p>\n

How the multilingual sign language model was built<\/p>\n

SL2T was trained on a dataset of more than 100,000 hours spanning over 50 sign languages, with roughly a quarter of that data specifically in ASL. Training the model across many languages and proficiency levels at once, rather than building separate single-language systems, helped it learn shared structures that improved overall performance \u2014 outperforming single-language models in DeepMind\u2019s own testing.<\/p>\n

That scale shows up in the numbers. On the FLEURS-ASL benchmark, which measures ASL-to-English translation quality, SL2T posted a zero-shot score of 70 BLEURT \u2014 a figure DeepMind says is significantly higher than anything previously reported for this kind of task. That gap matters: it signals SL2T isn\u2019t just an incremental improvement but a step change in how well AI can understand sign language as a full language, rather than as a series of hand gestures mapped to words.<\/p>\n

Part of the difficulty is that sign languages aren\u2019t simply spoken language performed with hands. They carry their own grammar, vocabulary, and structure, and they convey meaning through simultaneous movement of the hands, arms, torso, head, and face at once. Tracking all of that in real time is a demanding computer vision problem, and it\u2019s one reason earlier attempts \u2014 like sign language gloves \u2014 struggled, since they treated signing as a narrow hand-tracking task rather than genuine language translation.<\/p>\n

Pose landmarks protect signer privacy<\/p>\n

Rather than sending raw video to a server, SL2T processes signing as a stream of pose landmark coordinates<\/a>. An on-device model called MediaPipe Holistic<\/a> tracks points on the signer\u2019s body, and only those geometric coordinates \u2014 not the camera footage itself \u2014 get sent off for translation. The original video is discarded immediately, which is DeepMind\u2019s way of building privacy protection directly into the system\u2019s architecture rather than treating it as an afterthought.<\/p>\n

Skipping glosses for direct translation accuracy<\/p>\n

SL2T also breaks from a common approach in earlier sign language research: it translates coordinate sequences straight into text, bypassing intermediate step-by-step annotations known as \u201cglosses.\u201d Glosses tend to miss the richer, non-linear parts of sign languages, including facial expressions and spatial constructions that carry real meaning. By skipping that step, SL2T avoids artificial vocabulary limits and lets translation quality scale directly with more data.<\/p>\n

Google says it also worked on practical, real-world issues beyond benchmark scores \u2014 cutting down streaming delay, avoiding false translations when someone isn\u2019t actually signing, and specifically improving accuracy for the roughly 10% of signers who are left-handed, as well as for people signing one-handed while holding a phone in their other hand.<\/p>\n

Deaf community shaped SL2T\u2019s development<\/p>\n

DeepMind frames this as a project built with the Deaf community<\/a> rather than simply for it. The idea originated with Sam Sepah, a Deaf Googler, and Deaf partners were involved through data collection, user testing, and evaluation of the technology\u2019s real-world impact.<\/p>\n

To keep that involvement structured beyond the initial launch, Google formed the AI Sign Language Advisory Committee (AISLAC), bringing together global Deaf organizations and subject-matter experts to help steer how the technology gets deployed going forward. The company also co-authored a joint impact report alongside community partners for this first release, laying out both what SL2T can do and where its current limits are \u2014 an approach Google says it intends to repeat for future sign language releases.<\/p>\n

What\u2019s next for sign language AI translation<\/p>\n

DeepMind is positioning ASL support as a starting point rather than a finish line. The team says it\u2019s working to extend SL2T to additional sign languages, explore sign language generation \u2014 essentially the reverse process of turning text into signed output \u2014 and add more advanced AI capabilities on top of the existing model.<\/p>\n

The broader ambition ties back to Google\u2019s long-stated goal of making information universally accessible. Reaching that for Deaf communities means getting sign languages closer to parity with spoken and written languages across digital products, not just shipping one feature on one device. Whether SL2T\u2019s early ASL results hold up as it expands to dozens more sign languages, each with its own grammar and regional variation, will likely shape how quickly that broader vision becomes real for users beyond the Pixel 11.<\/p>\n

FAQ
\nWhat is SL2T and what does it do?<\/p>\n

SL2T is Google DeepMind\u2019s multilingual sign-language-to-text model. It translates signing directly into written text, letting Deaf users sign to their phone to type messages, search, or communicate instead of using a keyboard.<\/p>\n

Which sign languages does SL2T support currently?<\/p>\n

Right now, SL2T only supports American Sign Language translated into English, and it\u2019s available through Gboard and Live Transcribe on the Pixel 11. Additional sign languages are planned for future releases.<\/p>\n

How does SL2T protect user privacy during sign language translation?<\/p>\n

SL2T runs an on-device tracking system that reads points on a signer\u2019s body rather than sending raw video anywhere. Only those coordinates are used for translation, and the camera footage itself is discarded right away.<\/p>\n

How is the Deaf community involved in the SL2T project?<\/p>\n

Deaf individuals were involved from the earliest concept stage through data collection, testing, and governance. Google also set up the AI Sign Language Advisory Committee, made up of Deaf organizations and experts, to help guide how the technology continues to develop.<\/p>\n

Article produced with the assistance of artificial intelligence and reviewed by the editorial team.<\/p>\n","protected":false},"excerpt":{"rendered":"Google DeepMind has rolled out a new form of sign language AI translation that lets Deaf and hard…\n","protected":false},"author":2,"featured_media":138909,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[9],"tags":[24,5044,132,7543,2739,13367,3929],"class_list":["post-138908","post","type-post","status-publish","format-standard","has-post-thumbnail","category-google","tag-ai","tag-deepmind","tag-google","tag-google-deepmind","tag-language","tag-sign","tag-translation"],"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/138908","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=138908"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/138908\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/138909"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=138908"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=138908"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=138908"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}