AI RESEARCH
SignDPO: Multi-level Direct Preference Optimisation for Skeleton-based Gloss-free Sign Language Translation
arXiv CS.CL
•
ArXi:2604.18034v1 Announce Type: new We present SignDPO, a novel multi-level Direct Preference Optimisation (DPO) framework designed to enhance the alignment of skeleton-based Sign Language Translation. While current skeleton-based models have made significant progress using Maximum Likelihood Estimation, they are primarily constrained by an imitation-based paradigm that lacks discriminative sensitivity to the fine-grained spatio-temporal nuances of sign language, often leading to semantic drift.