Iterative Counterfactual Data Augmentation

TL;DR

This paper introduces an iterative counterfactual data augmentation method that reduces noise and spurious signals in datasets, leading to more accurate and human-aligned model explanations.

Contribution

The work presents a novel iterative CDA approach that converges to lower-noise datasets, improving the alignment of model rationales with human annotations.

Findings

Reduced spurious signals in augmented datasets

Improved alignment of model rationales with human annotations

Effective across multiple human and LLM-generated datasets

Abstract

Counterfactual data augmentation (CDA) is a method for controlling information or biases in training datasets by generating a complementary dataset with typically opposing biases. Prior work often either relies on hand-crafted rules or algorithmic CDA methods which can leave unwanted information in the augmented dataset. In this work, we show iterative CDA (ICDA) with initial, high-noise interventions can converge to a state with significantly lower noise. Our ICDA procedure produces a dataset where one target signal in the training dataset maintains high mutual information with a corresponding label and the information of spurious signals are reduced. We show training on the augmented datasets produces rationales on documents that better align with human annotation. Our experiments include six human produced datasets and two large-language model generated datasets.