Improving Lexically Constrained Neural Machine Translation with Source-Conditioned Masked Span Prediction

Open in new window