@misc{indiciae3933d433bbe7,
  title = {Attention Weights in Transformer NMT Fail Aligning Words Between Sequences but Largely Explain Model Predictions},
  author = {Javier Ferrando and Marta R. Costa-jussà},
  year = {2021},
  url = {https://arxiv.org/abs/2109.05853},
  note = {Source identifier: 2109.05853}
}