Google Deploys SL2T for ASL Dictation
Google DeepMind introduced its sign-language-to-text, or SL2T, model on August 12, bringing American Sign Language-to-English dictation to Gboard and Live Transcribe on Pixel 11. According to Google DeepMind, the model is initially available for ASL and English, with additional devices and sign languages to follow. SiliconANGLE reports that Google trained SL2T on more than 100,000 hours of sign-language data spanning more than 50 languages.
Google DeepMind introduced SL2T, a sign-language-to-text model that powers new ASL-to-English dictation features in Gboard and Live Transcribe on the Pixel 11. The August 12 release lets users sign into a phone camera where they would otherwise type, according to the Google DeepMind announcement.
The initial deployment supports American Sign Language (ASL) to English. Google DeepMind said the feature can be used for web searches, drafting messages and documents, and submitting queries or tasks to Gemini. In Live Transcribe, users can sign responses during conversations rather than typing them, the company said.
Google DeepMind describes SL2T as a "massively multilingual" translation model and said more devices and languages are forthcoming. The company also characterized the deployment as its first sign-language AI release in consumer products.
Training across sign languages
SiliconANGLE reports that Google trained SL2T on more than 100,000 hours of sign-language data covering over 50 languages, with ASL accounting for roughly one-quarter of the dataset. Google told the publication that multilingual training was intended to help the model learn structural patterns shared across sign languages, although the initial product release only translates ASL into English.
Google DeepMind notes that more than 200 sign languages are used worldwide and cites an estimated 70 million Deaf and hard-of-hearing sign-language users. Those figures underline a key technical constraint for the field: high-quality sign-language systems need to account for visual-spatial grammar, hand shape and motion, facial expression, and regional language variation rather than treating signing as a simple gesture-to-word mapping.
Product and engineering implications
SiliconANGLE reports that SL2T uses an on-device computer-vision component for sign tracking, though the excerpted report does not specify its architecture or the device-level latency, accuracy, and privacy characteristics. Google DeepMind's announcement likewise does not provide benchmark results, error rates, or a detailed language rollout schedule.
For ML practitioners, this release is notable because it moves sign-language recognition from a research setting into system input surfaces already used for speech dictation and transcription. Comparable accessibility deployments commonly place substantial importance on evaluation with native signers, robustness to camera framing and lighting, and transparent handling of language and dialect coverage. Public performance data will be important for assessing how broadly the initial ASL deployment generalizes across real-world signing conditions.
Key Points
- 1Google DeepMind deployed SL2T in Gboard and Live Transcribe, making ASL-to-English signing a consumer-device input method.
- 2Google reportedly trained SL2T on 100,000-plus hours across 50-plus sign languages, while initial product support remains ASL only.
- 3Consumer sign-language systems typically require rigorous native-signer evaluation because visual conditions, regional variation, and grammar affect recognition quality.
Scoring Rationale
SL2T is a notable real-world deployment of multilingual sign-language recognition in widely used mobile input products. Its initial ASL-only scope and absent public benchmark detail limit immediate general-purpose adoption assessment, but the release is important for accessibility-focused ML and on-device vision practitioners.
Sources
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
