Visualize: Figure 1 & 2 & 3.
Description: "An interpretation generated using AllenNLP
Interpret for NER. The model predicts three tags for an
input (top). We interpret each tag separately, e.g., input
reduction (bottom) removes as many
words as possible without changing a tag’s prediction.
Input reduction shows that the words “named”, “at”,
and “in downtown” are sufficient to predict the People, Organization, and Location tags, respectively.."