Interpretation of NLP models through input marginalization
arXiv:2010.13984
Abstract
To demystify the "black box" property of deep neural networks for natural language processing (NLP), several methods have been proposed to interpret their predictions by measuring the change in prediction probability after erasing each token of an input. Since existing methods replace each token with a predefined value (i.e., zero), the resulting sentence lies out of the training data distribution, yielding misleading interpretations. In this study, we raise the out-of-distribution problem induced by the existing interpretation methods and present a remedy; we propose to marginalize each token out. We interpret various NLP models trained for sentiment analysis and natural language inference using the proposed method.
10 pages, 5 figures, to be published in the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP 2020)
References in corpus (7)
- Attention is not Explanation
- Understanding Neural Networks through Representation Erasure
- Visualizing Deep Neural Network Decisions: Prediction Difference Analysis
- Pathologies of Neural Models Make Interpretations Difficult
- Visualizing and Understanding Neural Models in NLP
- Towards Hierarchical Importance Attribution: Explaining Compositional Semantics for Neural Sequence Models
- Information-Theoretic Visual Explanation for Black-Box Classifiers