Publications

On the strengths of cross-attention in pretrained transformers for machine translation