New AI system improves sign language generation by capturing facial expressions and body movements that are grammatically essential

Researchers introduced SMPL-FX, which combines FLAME's facial expression capabilities with SMPL-X body model to better represent non-manual features like mouthings, eyebrow raises, and head movements

M3T, an autoregressive transformer, uses modality-specific tokenization for body, hands, and face to avoid codebook collapse that limits expression in previous systems

The system addresses a key limitation: standard body models lack sufficient facial detail to encode grammatically required sign language features that cannot be recovered from hand movements alone