When Distributions Shifts: Causal Generalization for Low-Resource Languages

TL;DR

This paper explores causal domain generalization methods to improve NLP model robustness in low-resource languages under distribution shifts, using data augmentation and invariant representation learning.

Contribution

It introduces two causal DG approaches—counterfactual data augmentation with GPT-4o-mini and invariant causal representation learning with DINER—adapted for low-resource multilingual NLP.

Findings

Counterfactual augmentation improves domain robustness.

Causal representation learning enhances out-of-distribution performance.

Methods show consistent gains across multiple low-resource languages.

Abstract

Machine learning models often fail under distribution shifts, a problem exacerbated in low-resource settings where limited data restricts robust generalization. Domain generalization(DG) methods address this challenge by learning representations that remain invariant across domains, frequently leveraging causal principles. In this work, we study two causal DG approaches for low-resource natural language processing. First, we apply causal data augmentation using GPT-4o-mini to generate counterfactual paraphrases for sentiment classification on the NaijaSenti Twitter corpus in Yoruba and Igbo. Second, we investigate invariant causal representation learning with the Debiasing in Aspect Review (DINER) framework for aspect-based sentiment analysis. We extend DINER to a multilingual setting by introducing Afri-SemEval, a dataset of 17 languages translated from SemEval-2014 Task. Experiments show improved robustness to unseen domains, with consistent gains from counterfactual augmentation and enhanced out-of-distribution performance from causal representation learning across multiple languages.