Consistent Document-Level Relation Extraction via Counterfactuals

TL;DR

This paper introduces CovEReD, a counterfactual data generation method for document-level relation extraction, which improves model consistency by reducing reliance on spurious signals without harming overall performance.

Contribution

The paper presents a novel counterfactual data augmentation approach for document-level RE that enhances model consistency and reduces bias from spurious correlations.

Findings

Models trained on factual data are inconsistent after entity replacement.

Counterfactual training maintains consistency with minimal performance loss.

Re-DocRED-CF dataset enables evaluation of model robustness against biases.

Abstract

Many datasets have been developed to train and evaluate document-level relation extraction (RE) models. Most of these are constructed using real-world data. It has been shown that RE models trained on real-world data suffer from factual biases. To evaluate and address this issue, we present CovEReD, a counterfactual data generation approach for document-level relation extraction datasets using entity replacement. We first demonstrate that models trained on factual data exhibit inconsistent behavior: while they accurately extract triples from factual data, they fail to extract the same triples after counterfactual modification. This inconsistency suggests that models trained on factual data rely on spurious signals such as specific entities and external knowledge $\unicode{x2013}$ rather than on the input context $\unicode{x2013}$ to extract triples. We show that by generating document-level counterfactual data with CovEReD and training models on them, consistency is maintained with minimal impact on RE performance. We release our CovEReD pipeline as well as Re-DocRED-CF, a dataset of counterfactual RE documents, to assist in evaluating and addressing inconsistency in document-level RE.