DISCO: Distilling Counterfactuals with Large Language Models
This work addresses the problem of limited and expensive counterfactual data generation for machine learning practitioners, offering a scalable solution that enhances model robustness and generalization, though it is incremental as it builds on existing data augmentation and distillation techniques.
The authors tackled the scarcity of high-quality counterfactual data for robust generalization by introducing DISCO, a method that automatically generates such data at scale using large language models and task-specific filtering, resulting in improved robustness (6% absolute gain), better cross-distribution generalization (2%), and 10% higher consistency in models trained on NLI tasks.
Models trained with counterfactually augmented data learn representations of the causal structure of tasks, enabling robust generalization. However, high-quality counterfactual data is scarce for most tasks and not easily generated at scale. When crowdsourced, such data is typically limited in scale and diversity; when generated using supervised methods, it is computationally expensive to extend to new counterfactual dimensions. In this work, we introduce DISCO (DIStilled COunterfactual Data), a new method for automatically generating high quality counterfactual data at scale. DISCO engineers prompts to generate phrasal perturbations with a large general language model. Then, a task-specific teacher model filters these generations to distill high-quality counterfactual data. While task-agnostic, we apply our pipeline to the task of natural language inference (NLI) and find that on challenging evaluations such as the NLI stress test, comparatively smaller student models trained with DISCO generated counterfactuals are more robust (6% absolute) and generalize better across distributions (2%) compared to models trained without data augmentation. Furthermore, DISCO augmented models are 10% more consistent between counterfactual pairs on three evaluation sets, demonstrating that DISCO augmentation enables models to more reliably learn causal representations. Our repository is available at: https://github.com/eric11eca/disco