NeuroCounterfactuals: Beyond Minimal-Edit Counterfactuals for Richer Data Augmentation

TL;DR

NeuroCounterfactuals introduces a generative method for creating rich, naturalistic counterfactual data with larger edits, improving sentiment classification robustness beyond minimal-edit approaches.

Contribution

The paper proposes NeuroCounterfactuals, a novel generative approach that produces diverse, naturalistic counterfactuals with larger edits for data augmentation in NLP.

Findings

Outperforms manual counterfactuals in sentiment classification tasks.

Enhances in-domain and out-of-domain model robustness.

Provides detailed analysis of advantages over minimal-edit methods.

Abstract

While counterfactual data augmentation offers a promising step towards robust generalization in natural language processing, producing a set of counterfactuals that offer valuable inductive bias for models remains a challenge. Most existing approaches for producing counterfactuals, manual or automated, rely on small perturbations via minimal edits, resulting in simplistic changes. We introduce NeuroCounterfactuals, designed as loose counterfactuals, allowing for larger edits which result in naturalistic generations containing linguistic diversity, while still bearing similarity to the original document. Our novel generative approach bridges the benefits of constrained decoding, with those of language model adaptation for sentiment steering. Training data augmentation with our generations results in both in-domain and out-of-domain improvements for sentiment classification, outperforming even manually curated counterfactuals, under select settings. We further present detailed analyses to show the advantages of NeuroCounterfactuals over approaches involving simple, minimal edits.