Mitigating Annotation Artifacts in Natural Language Inference Datasets to Improve Cross-dataset Generalization Ability

TL;DR

This paper investigates annotation artifacts in NLI datasets that bias models and hinder cross-dataset generalization, proposing a training framework to mitigate these artifacts and improve model robustness.

Contribution

It introduces a novel training framework designed to reduce annotation artifacts in NLI datasets, enhancing cross-dataset generalization performance.

Findings

Mitigation of annotation artifacts improves cross-dataset NLI accuracy.

Proposed methods reduce bias and enhance model robustness.

Experimental results show significant generalization gains.

Abstract

Natural language inference (NLI) aims at predicting the relationship between a given pair of premise and hypothesis. However, several works have found that there widely exists a bias pattern called annotation artifacts in NLI datasets, making it possible to identify the label only by looking at the hypothesis. This irregularity makes the evaluation results over-estimated and affects models' generalization ability. In this paper, we consider a more trust-worthy setting, i.e., cross-dataset evaluation. We explore the impacts of annotation artifacts in cross-dataset testing. Furthermore, we propose a training framework to mitigate the impacts of the bias pattern. Experimental results demonstrate that our methods can alleviate the negative effect of the artifacts and improve the generalization ability of models.