Visualize: Figure1 & 2 & 3.
Description: " Comparison of our saliency-based word
alignment interpretation of convolutional NMT model
with reference and attention interpretation"