Explaining Natural Language Processing Classifiers with Occlusion and Language Modeling

Abstract

Deep neural networks are powerful statistical learners. However, their predictions do not come with an explanation of their process. To analyze these models, explanation methods are being developed. We present a novel explanation method, called OLM, for natural language processing classifiers. This method combines occlusion and language modeling, which are techniques central to explainability and NLP, respectively. OLM gives explanations that are theoretically sound and easy to understand. We make several contributions to the theory of explanation methods. Axioms for explanation methods are an interesting theoretical concept to explore their basics and deduce methods. We introduce a new axiom, give its intuition and show it contradicts another existing axiom. Additionally, we point out theoretical difficulties of existing gradient-based and some occlusion-based explanation methods in natural language processing. We provide an extensive argument why evaluation of explanation methods is difficult. We compare OLM to other explanation methods and underline its uniqueness experimentally. Finally, we investigate corner cases of OLM and discuss its validity and possible improvements.

Master's Thesis at University of Potsdam without Acknowledgements

References in corpus (15)

• Efficient Estimation of Word Representations in Vector Space
• Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation
• Axiomatic Attribution for Deep Networks
• Methods for Interpreting and Understanding Deep Neural Networks
• Learning Important Features Through Propagating Activation Differences
• European Union regulations on algorithmic decision-making and a "right to explanation"
• Opening the Black Box of Deep Neural Networks via Information
• How to Explain Individual Classification Decisions
• The Loss Surfaces of Multilayer Networks
• Sanity Checks for Saliency Maps
• Learning how to explain neural networks: PatternNet and PatternAttribution
• "What is Relevant in a Text Document?": An Interpretable Machine Learning Approach
• End-to-end ASR: from Supervised to Semi-Supervised Learning with Modern Architectures
• Full-Gradient Representation for Neural Network Visualization
• A Human-Grounded Evaluation Benchmark for Local Explanations of Machine Learning