Q-BridgeNet unifies multilingual sign language translation

Researchers propose Q-BridgeNet, a unified framework for multilingual sign language translation that jointly mitigates cross-lingual conflicts on both the sign and spoken language sides.

• On the sign side, it learns discrete Q-units via adaptive segmentation and residual vector quantization using shared and language-specific codebooks.
• On the spoken side, a multilingual LLM is fine-tuned to operate in the Q-unit space leveraging cross-lingual priors.
• Experiments on PHOENIX14T, How2Sign, and CSL-Daily show state-of-the-art performance on native pairs and strong generalization to non-native pairs.

The approach enables communication across diverse sign and spoken language communities by capturing shared semantics while preserving language-specific variations.