Google DeepMind’s SL2T Model Brings Sign-to-Text Dictation to the Pixel 11

Google DeepMind on August 12, 2026, launched SL2T, a sign-language-to-text model that lets Deaf and hard of hearing users sign to their phone anywhere they would normally type. The model debuts alongside Google’s new Pixel 11 family, where it powers sign-to-text dictation inside Gboard and Live Transcribe, starting with American Sign Language (ASL) to English translation.

Voice-to-text dictation has been a smartphone staple for well over a decade, but according to DeepMind’s announcement, that same convenience has never reached the world’s more than 200 sign languages or the estimated 70 million Deaf and hard of hearing people who use them. DeepMind says SL2T marks the first time it is bringing sign language AI out of the lab and into consumer products.

How SL2T Works

SL2T is a massively multilingual model trained on more than 100,000 hours of sign language data spanning over 50 languages, with roughly a quarter of that dataset in ASL, according to DeepMind. Training the model jointly across many sign languages, dialects, and proficiency levels, rather than building a separate model per language, let it learn structures shared across sign languages and outperform single-language models in DeepMind’s own testing. On the FLEURS-ASL benchmark, DeepMind reports SL2T reaches a zero-shot score of 70 BLEURT, which the company describes as significantly higher than any previously reported score on that test.

The model never processes raw video. An on-device computer vision model called MediaPipe Holistic first converts a signer’s hand, arm, torso, head, and facial movements into a stream of pose landmark coordinates, and only those coordinates, not camera footage, are sent to Google’s servers for translation; DeepMind says the original video is discarded on-device immediately afterward. The company frames this as a deliberate privacy safeguard.

SL2T also breaks from how earlier sign-language-to-text systems worked. Prior systems typically translated signs into an intermediate representation called glosses before converting that into spoken-language text. DeepMind says, “glosses fail to capture rich, non-linear aspects of sign languages such as non-manual markers and spatial constructions,” so SL2T instead translates landmark sequences directly into text. “Translating directly from landmarks removes artificial vocabulary limits and allows translation quality to scale directly with data,” the company said.

What It Looks Like on the Pixel 11

On the Pixel 11 Pro Fold, users open the Translate app or tap a sign language icon in Gboard’s toolbar, then sign in front of the phone’s front camera, according to Android Authority, which tested the feature at launch. The transcription appears on screen, and on the Pro Fold it also mirrors to the cover screen so the other person in the conversation can read it and respond. In Live Transcribe, users can sign their side of a conversation instead of typing back and forth, and the same sign input works for web searches, messages, documents, and queries to Gemini.

The launch is deliberately narrow. SL2T supports only ASL-to-English translation for now, and Android Authority reported the feature is available in select countries and regions to start, on the Pixel 11 series specifically. Engadget, which also covered the launch, reported that Google plans to bring the feature to more devices and add more sign languages over time, though neither company has committed to a timeline.

Built With Deaf Advisors, Not Just for Them

DeepMind said the project began with Sam Sepah, a Deaf Googler, and involved Deaf partners at every stage that followed, including data collection, user studies, and impact assessment. To guide the rollout, Google formed the AI Sign Language Advisory Committee (AISLAC), bringing together Deaf organizations and subject-matter experts from around the world, and the company co-authored a joint impact report describing SL2T’s capabilities and current limitations alongside the release.

DeepMind is upfront that the model is not finished work. The company acknowledges errors persist on rare signs, rapid fingerspelling, passive constructions, and tense cues that depend on context, the kinds of edge cases that tend to be hardest for any translation system trained mostly on common usage.

Why It Matters

SL2T ships as a narrow first release: one language pair, one phone family, a limited set of countries. But it is also a concrete test of whether an AI lab can move a technology that has mostly lived in research papers and pilot programs into a mainstream consumer product without treating it as an afterthought. DeepMind frames its goal as reaching full parity between how sign languages and spoken or written languages are served by AI, a bar the company has only begun to clear with this release. Whether SL2T expands quickly to the dozens of other sign languages already represented in its training data, or stalls the way earlier accessibility features sometimes have, will say more about Google’s follow-through than today’s announcement can.