Google Unveils SL2T AI Model for Real-Time Sign Language Translation

Google DeepMind has introduced SL2T, a sign-language-to-text (SL2T) AI model that powers real-time sign language features for Deaf and hard-of-hearing users. The technology debuts on Gboard and Live Transcribe for Pixel 11, starting with American Sign Language (ASL) to English translations. More sign languages and devices are expected soon, according to a blog announcement on August 12, 2026.

SL2T addresses an often-overlooked gap in accessibility. While AI has transformed spoken language processing, it has largely bypassed the estimated 70 million people who use one of the world’s 200+ sign languages. Traditional challenges, like the complexity of sign languages and their reliance on simultaneous physical movements, have made machine translation difficult. SL2T aims to overcome these hurdles by leveraging massive multilingual data and a novel, user-centric design.

How SL2T Works and Why It Stands Out

The model processes over 100,000 hours of data spanning 50+ sign languages, with ASL comprising about 25% of the dataset. Unlike earlier approaches that relied on intermediate annotations known as “glosses,” SL2T translates directly from pose landmark locations to text. This sidesteps limitations in vocabulary and grammar, delivering more natural results. On benchmarks like FLEURS-ASL, SL2T achieved a zero-shot BLEURT score of 70, significantly outperforming prior models.

Privacy is a key focus. SL2T doesn’t transmit raw video feeds to servers. Instead, it extracts geometric coordinates of a signer’s movements using an on-device model called MediaPipe Holistic. These coordinates are then used for translation, ensuring user video data is discarded immediately.

Real-World Applications

SL2T allows Deaf users to sign directly to their phones for tasks like web searches, messaging, and document drafting, offering a faster and more natural alternative to typing. For conversational use, Live Transcribe enables users to respond in sign language rather than typing, bridging communication gaps between Deaf and hearing communities. Early testers reported that signing in ASL felt more intuitive and efficient than typing in English.

Challenges and Next Steps

Despite its advancements, SL2T is not without limitations. It struggles with rare signs, rapid finger spelling, and ambiguous grammatical constructions. Context-dependent tasks like tense interpretation also remain challenging. Google has addressed practical issues such as minimizing latency, ensuring fairness for left-handed users, and improving performance for one-handed signing.

Google DeepMind emphasized its commitment to collaboration with the Deaf community. The company established the AI Sign Language Advisory Committee (AISLAC) to guide ethical deployment and co-authored an impact report with Deaf experts. This participatory approach aims to ensure the technology meets the needs of its target users.

Looking Ahead

While SL2T’s launch is a significant milestone, it represents just the beginning of Google’s ambitions in accessibility. The team plans to expand SL2T to additional sign languages and explore sign language generation capabilities. The ultimate goal is to achieve parity between sign language processing and spoken/written language AI. For now, users can experience SL2T on Pixel 11 at no additional cost, with broader device support expected in the near future.

Google’s initiative underscores a broader trend of AI moving beyond traditional language models to address underserved communities. As accessibility becomes a key focus in AI development, technologies like SL2T could set a precedent for inclusive innovation.

Image source: Shutterstock Source