Target Adaptive Training

AbstractThe decoder in simultaneous neural machine translation receives limited information from the source while having to balance the opposing requirements of latency versus translation quality. In this paper, we use …...