Evaluating Models’ Local Decision Boundaries via Contrast Sets

By Matt Gardner, Yoav Artzi, Jonathan Berant, Ben Bogin, Sihao Chen, Dheeru Dua, Yanai Elazar, Ananth Gottumukkala, Nitish Gupta, Hannaneh Hajishirzi and 13 more. 2020

462 citations recorded by the source corpus.

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 the abilities a dataset is intended to test. We propose a more rigorous 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, and 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.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Evaluation & Benchmarks

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Evaluation & Benchmarks

Citations in the corpus

UnifiedQA: Crossing Format Boundaries With a Single QA System (2020, methodology): “Additionally, we use contrast-sets (Gardner et al., 2020) for several of our datasets (denoted with “CS”): BoolQ-CS, ROPES-CS, Quoref-CS, DROP-CS.”
Are NLP Models really able to Solve Simple Math Word Problems? (2021, background): “Gardner et al. (2020) suggested creating contrast sets by manually perturbing test instances in small yet meaningful ways that change the gold label.”
HateCheck: Functional Tests for Hate Speech Detection Models (2020, background): “constructed to be particularly challenging to models relying on overly simplistic decision rules and thus enable more accurate testing of true model functionalities (Gardner et al., 2020).”
Ethical-Advice Taker: Do Language Models Understand Natural Language Interventions? (2021, background): “A range of recent works are based on the general idea of models revising their behavior according to changes in their input (Wallace et al., 2019; Gardner et al., 2020; Emelin et al., 2020; Ye and Ren, 2021; Schick and Schütze, 2020; Sheng et al., 2020).”
CommonsenseQA 2.0: Exposing the Limits of AI through Gamification (2021, background): “Contrast sets To examine the consistency of our baselines, we created contrast sets [8, 9], that is, we took 60 questions from the development set and authored additional questions that are minor perturbations of the original question.”
Tailor: Generating and Perturbing Text with Semantic Controls (2021, methodology): “Because the training objective of our generator does not provide deterministic signal for (Gardner et al., 2020; Ribeiro et al., 2019). noun attachment outputs, we do not expect our verb → noun strategy to always result in generations with noun attachment.”
Data and its (dis)contents: A survey of dataset development and use in machine learning research (2020, methodology): “Recent papers also outline methodologies for leveraging human insight in the manual construction of counterfactual examples that complement instances in natural language processing datasets to promote better generalization [Gardner et al., 2020, Kaushik et al., 2020].”
What Will it Take to Fix Benchmarking in Natural Language Understanding? (2021, background): “The Build-it-Break-it challenge (Ettinger et al., 2017), the Open Reading Benchmark (Dua et al., 2019), and the Gardner et al. (2020) contrast sets, among their other features, allow expert annotators to add examples to a test set to ﬁll perceived gaps in coverage or correct perceived artifacts in…”
Learning to Solve Complex Tasks by Talking to Agents (2021, background): “Despite this progress, existing benchmarks often contain many shortcuts (Trivedi et al., 2020; Min et al., 2019a), resulting in models being brittle (Gardner et al., 2020).”