Single-Sentence Reader: A Novel Approach for Addressing Answer Position Bias

TL;DR

This paper introduces a Single-Sentence Reader approach to mitigate answer position bias in machine reading comprehension, improving model robustness by reducing reliance on spurious answer location cues.

Contribution

The paper proposes a novel Single-Sentence Reader method that effectively addresses answer position bias in MRC models, demonstrating near-normal performance on biased datasets.

Findings

Single-Sentence Readers perform well on biased datasets

Models trained with this approach nearly match normal dataset performance

The method enhances robustness against answer position bias

Abstract

Machine Reading Comprehension (MRC) models tend to take advantage of spurious correlations (also known as dataset bias or annotation artifacts in the research community). Consequently, these models may perform the MRC task without fully comprehending the given context and question, which is undesirable since it may result in low robustness against distribution shift. The main focus of this paper is answer-position bias, where a significant percentage of training questions have answers located solely in the first sentence of the context. We propose a Single-Sentence Reader as a new approach for addressing answer position bias in MRC. Remarkably, in our experiments with six different models, our proposed Single-Sentence Readers trained on biased dataset achieve results that nearly match those of models trained on normal dataset, proving their effectiveness in addressing the answer position bias. Our study also discusses several challenges our Single-Sentence Readers encounter and proposes a potential solution.