On the Sensitivity and Stability of Model Interpretations in NLP
This work addresses the challenge of evaluating interpretation faithfulness in NLP, which is crucial for researchers and practitioners relying on model interpretability, though it is incremental as it builds on existing removal-based criteria.
The paper tackles the problem of defining and measuring faithfulness in NLP model interpretations by proposing two new criteria, sensitivity and stability, and shows that conclusions about faithfulness vary substantially based on different notions. It introduces a new class of interpretation methods using adversarial robustness techniques, which are effective under the new criteria and overcome limitations of gradient-based methods, with applications in text classification and dependency parsing.
Recent years have witnessed the emergence of a variety of post-hoc interpretations that aim to uncover how natural language processing (NLP) models make predictions. Despite the surge of new interpretation methods, it remains an open problem how to define and quantitatively measure the faithfulness of interpretations, i.e., to what extent interpretations reflect the reasoning process by a model. We propose two new criteria, sensitivity and stability, that provide complementary notions of faithfulness to the existed removal-based criteria. Our results show that the conclusion for how faithful interpretations are could vary substantially based on different notions. Motivated by the desiderata of sensitivity and stability, we introduce a new class of interpretation methods that adopt techniques from adversarial robustness. Empirical results show that our proposed methods are effective under the new criteria and overcome limitations of gradient-based methods on removal-based criteria. Besides text classification, we also apply interpretation methods and metrics to dependency parsing. Our results shed light on understanding the diverse set of interpretations.