Video-based Sign Language Recognition without Temporal Segmentation