Visualize: Figure1 & 2 & 5.
Description: "we propose to interpret the intermediate layers of NLI models by visualizing the saliency of attention and LSTM gating signals."