DeepMind SL2T brings sign-to-text phones

SL2T is a meaningful accessibility milestone because it treats sign language as a full visual language, not merely a sequence of hand gestures.

• –The model accounts for hands, body movement, facial expression, and spatial context
• –On-device pose tracking can improve privacy by discarding raw video before translation
• –Direct sign-to-text translation avoids brittle intermediate gloss representations
• –Initial ASL-to-English support is useful, but broader language coverage will determine its global impact
• –The Pixel-first rollout shows how specialized multimodal models can reach users through existing mobile interfaces