Evaluating Models' Local Decision Boundaries via Contrast Sets

TL;DR

This paper introduces contrast sets, a new annotation method involving small perturbations to test data, to better evaluate models' true linguistic understanding beyond standard test sets.

Contribution

It proposes a systematic annotation paradigm for creating contrast sets that reveal models' local decision boundaries and true capabilities.

Findings

Model performance drops up to 25% on contrast sets

Contrast sets reveal systematic gaps in models' understanding

The approach improves evaluation of linguistic capabilities

Abstract

Standard test sets for supervised learning evaluate in-distribution generalization. Unfortunately, when a dataset has systematic gaps (e.g., annotation artifacts), these evaluations are misleading: a model can learn simple decision rules that perform well on the test set but do not capture a dataset's intended capabilities. We propose a new annotation paradigm for NLP that helps to close systematic gaps in the test data. In particular, after a dataset is constructed, we recommend that the dataset authors manually perturb the test instances in small but meaningful ways that (typically) change the gold label, creating contrast sets. Contrast sets provide a local view of a model's decision boundary, which can be used to more accurately evaluate a model's true linguistic capabilities. We demonstrate the efficacy of contrast sets by creating them for 10 diverse NLP datasets (e.g., DROP reading comprehension, UD parsing, IMDb sentiment analysis). Although our contrast sets are not explicitly adversarial, model performance is significantly lower on them than on the original test sets---up to 25\% in some cases. We release our contrast sets as new evaluation benchmarks and encourage future dataset construction efforts to follow similar annotation processes.