Unsupervised Explanation Generation for Machine Reading Comprehension
This addresses the problem of model interpretability for users in real-life applications, though it is incremental as it builds on existing MRC methods.
The paper tackles the lack of explainability in machine reading comprehension models by proposing a self-explainable framework that filters passage information to generate explanations, achieving consistent improvements over baselines on three datasets and outperforming attention mechanisms in human evaluations.
With the blooming of various Pre-trained Language Models (PLMs), Machine Reading Comprehension (MRC) has embraced significant improvements on various benchmarks and even surpass human performances. However, the existing works only target on the accuracy of the final predictions and neglect the importance of the explanations for the prediction, which is a big obstacle when utilizing these models in real-life applications to convince humans. In this paper, we propose a self-explainable framework for the machine reading comprehension task. The main idea is that the proposed system tries to use less passage information and achieve similar results compared to the system that uses the whole passage, while the filtered passage will be used as explanations. We carried out experiments on three multiple-choice MRC datasets, and found that the proposed system could achieve consistent improvements over baseline systems. To evaluate the explainability, we compared our approach with the traditional attention mechanism in human evaluations and found that the proposed system has a notable advantage over the latter one.