H. Keleş

ODE-inspired update dynamics can significantly boost sign language translation performance without the need for larger models.

Weak supervision from loosely aligned broadcast transcripts can yield sign representations that significantly outperform traditional methods in both localization and translation tasks.

Ditch deterministic embeddings: A novel variational approach to sign language production achieves state-of-the-art back-translation and more realistic motion by modeling articulator-specific latent distributions.