Cross-modality Data Augmentation for End-to-End Sign Language Translation