DISCERN: Decoding Systematic Errors in Natural Language for Text Classifiers

TL;DR

DISCERN is a framework that uses language explanations generated by large language models to interpret, identify, and mitigate systematic biases in text classifiers, leading to improved performance and better human interpretability.

Contribution

We introduce DISCERN, a novel method that employs language explanations and an interactive loop between language models to interpret and reduce systematic biases in text classifiers.

Findings

Language explanations improve classifier performance beyond bias exemplars.

Users interpret systematic biases more effectively with language explanations.

Our framework achieves consistent performance gains across multiple datasets.

Abstract

Despite their high predictive accuracies, current machine learning systems often exhibit systematic biases stemming from annotation artifacts or insufficient support for certain classes in the dataset. Recent work proposes automatic methods for identifying and explaining systematic biases using keywords. We introduce DISCERN, a framework for interpreting systematic biases in text classifiers using language explanations. DISCERN iteratively generates precise natural language descriptions of systematic errors by employing an interactive loop between two large language models. Finally, we use the descriptions to improve classifiers by augmenting classifier training sets with synthetically generated instances or annotated examples via active learning. On three text-classification datasets, we demonstrate that language explanations from our framework induce consistent performance improvements that go beyond what is achievable with exemplars of systematic bias. Finally, in human evaluations, we show that users can interpret systematic biases more effectively (by over 25% relative) and efficiently when described through language explanations as opposed to cluster exemplars.