Multilingual Multi-Aspect Explainability Analyses on Machine Reading Comprehension Models

Abstract

Achieving human-level performance on some of the Machine Reading Comprehension (MRC) datasets is no longer challenging with the help of powerful Pre-trained Language Models (PLMs). However, the internal mechanism of these artifacts remains unclear, placing an obstacle for further understanding these models. This paper focuses on conducting a series of analytical experiments to examine the relations between the multi-head self-attention and the final MRC system performance, revealing the potential explainability in PLM-based MRC models. To ensure the robustness of the analyses, we perform our experiments in a multilingual way on top of various PLMs. We discover that passage-to-question and passage understanding attentions are the most important ones in the question answering process, showing strong correlations to the final performance than other parts. Through comprehensive visualizations and case studies, we also observe several general findings on the attention maps, which can be helpful to understand how these models solve the questions.

15 pages

References in corpus (6)

• Neural Machine Translation by Jointly Learning to Align and Translate
• TensorFlow: A system for large-scale machine learning
• Interpretable machine learning: definitions, methods, and applications
• Pre-Training with Whole Word Masking for Chinese BERT
• Captum: A unified and generic model interpretability library for PyTorch
• ExpMRC: Explainability Evaluation for Machine Reading Comprehension

Cited by in corpus (3)

• ExpMRC: Explainability Evaluation for Machine Reading Comprehension
• Evaluating Large Language Models on Multimodal Chemistry Olympiad Exams
• Visualizing attention zones in machine reading comprehension models