mlsanityData to 7 Oct 2026

Evaluating and Characterizing Human Rationales

arXiv 2010.04736 ↗First posted 9 Oct 2020cs.CL cs.AI cs.CYcited by 36

Not on the map. Why?

Abstract (arXiv)

Two main approaches for evaluating the quality of machine-generated rationales are: 1) using human rationales as a gold standard; and 2) automated metrics based on how rationales affect model behavior. An open question, however, is how human rationales fare with these automatic metrics. Analyzing a variety of datasets and models, we find that human rationales do not necessarily perform well on these metrics. To unpack this finding, we propose improved metrics to account for model-dependent baseline performance. We then propose two methods to further characterize rationale quality, one based on model retraining and one on using "fidelity curves" to reveal properties such as irrelevance and redundancy. Our work leads to actionable suggestions for evaluating and characterizing rationales.

Key facts. Cited by 36 papers in the map (data through 7 Oct 2026). First posted on arXiv 9 Oct 2020.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 11; 2022: 9; 2023: 6; 2024: 5; 2025: 4; 2026: 1.

36 papers.

Sort:

By area:

  1. 2021Teach Me to Explain: A Review of Datasets for Explainable Natural Language Processingcited by 126
  2. 2021Few-Shot Self-Rationalization with Natural Language Promptscited by 83
  3. 2021UNIREX: A Unified Learning Framework for Language Model Rationale Extractioncited by 32
  4. 2021Local Interpretations for Explainable Natural Language Processing: A Surveycited by 31
  5. 2021Is Sparse Attention more Interpretable?cited by 31
  6. 2021On the Diversity and Limits of Human Explanationscited by 30
  7. 2022ferret: a Framework for Benchmarking Explainers on Transformerscited by 26
  8. 2021What to Learn, and How: Toward Effective Learning from Rationalescited by 25
  9. 2022An Empirical Study on Explanations in Out-of-Domain Settingscited by 21
  10. 2023CATfOOD: Counterfactual Augmented Training for Improving Out-of-Domain Performance and Calibrationcited by 16
  11. 2022A survey on improving NLP models with human explanationscited by 15
  12. 2022On the Impact of Temporal Concept Drift on Model Explanationscited by 14
24 more
  1. 2023Incorporating Attribution Importance for Improving Faithfulness Metricscited by 14
  2. 2023Are Human Explanations Always Helpful? Towards Objective Evaluation of Human Natural Language Explanationscited by 13
  3. 2025GeoZero: Incentivizing Reasoning from Scratch on Geospatial Scenescited by 13
  4. 2021Unifying Model Explainability and Robustness for Joint Text Classification and Rationale Extractioncited by 11
  5. 2021Flexible Instance-Specific Rationalization of NLP Modelscited by 10
  6. 2023Beyond Labels: Empowering Human Annotators with Natural Language Explanations through a Novel Active-Learning Architecturecited by 10
  7. 2022Does Self-Rationalization Improve Robustness to Spurious Correlations?cited by 9
  8. 2022Are Shortest Rationales the Best Explanations for Human Understanding?cited by 8
  9. 2022XMD: An End-to-End Framework for Interactive Explanation-Based Debugging of NLP Modelscited by 8
  10. 2022ER-Test: Evaluating Explanation Regularization Methods for Language Modelscited by 6
  11. 2024Comparing Explanation Faithfulness between Multilingual and Monolingual Fine-tuned Language Modelscited by 6
  12. 2025Should I Share this Translation? Evaluating Quality Feedback for User Reliance on Machine Translationcited by 5
  13. 2021On Positivity Bias in Negative Reviewscited by 3
  14. 2024The Role of Syntactic Span Preferences in Post-Hoc Explanation Disagreementcited by 3
  15. 2024Counterfactuals As a Means for Evaluating Faithfulness of Attribution Methods in Autoregressive Language Modelscited by 3
  16. 2021You Can Do Better! If You Elaborate the Reason When Making Predictioncited by 2
  17. 2022Towards Human-Centred Explainability Benchmarks For Text Classificationcited by 2
  18. 2023Situated Natural Language Explanationscited by 2
  19. 2023Measuring Information in Text Explanationscited by 1
  20. 2024Evaluating Saliency Explanations in NLP by Crowdsourcingcited by 1
  21. 2024Explanation Regularisation through the Lens of Attributionscited by 1
  22. 2025Learning from Sufficient Rationales: Analysing the Relationship Between Explanation Faithfulness and Token-level Regularisation Strategiescited by 1
  23. 2025Can human clinical rationales improve the performance and explainability of clinical text classification models?cited by 0
  24. 2026Legal Experts Disagree With Rationale Extraction Techniques for Explaining ECtHR Case Outcome Classificationcited by 0