Towards Hierarchical Importance Attribution: Explaining Compositional Semantics for Neural Sequence Models

Open in new window