mlsanityData to 7 Oct 2026

The Probabilities Also Matter: A More Faithful Metric for Faithfulness of Free-Text Explanations in Large Language Models

arXiv 2404.03189 ↗First posted 4 Apr 2024cs.CL cs.AIcited by 29

Not on the map. Why?

Abstract (arXiv)

In order to oversee advanced AI systems, it is important to understand their underlying decision-making process. When prompted, large language models (LLMs) can provide natural language explanations or reasoning traces that sound plausible and receive high ratings from human annotators. However, it is unclear to what extent these explanations are faithful, i.e., truly capture the factors responsible for the model's predictions. In this work, we introduce Correlational Explanatory Faithfulness (CEF), a metric that can be used in faithfulness tests based on input interventions. Previous metrics used in such tests take into account only binary changes in the predictions. Our metric accounts for the total shift in the model's predicted label distribution, more accurately reflecting the explanations' faithfulness. We then introduce the Correlational Counterfactual Test (CCT) by instantiating CEF on the Counterfactual Test (CT) from Atanasova et al. (2023). We evaluate the faithfulness of free-text explanations generated by few-shot-prompted LLMs from the Llama2 family on three NLP tasks. We find that our metric measures aspects of faithfulness which the CT misses.

Key facts. Cited by 29 papers in the map (data through 7 Oct 2026). Builds on 29 of its 56 references that are in the map. First posted on arXiv 4 Apr 2024.

Builds on

29 of 56 references.

References · 29

  1. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 12,865
  2. 2018CommonsenseQA: A Question Answering Challenge Targeting Commonsense Knowledgecited by 1,793
  3. 2018Sanity Checks for Saliency Mapscited by 1,149
  4. 2023Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Promptingcited by 1,081
  5. 2022Discovering Latent Knowledge in Language Models Without Supervisioncited by 733
  6. 2023Measuring Faithfulness in Chain-of-Thought Reasoningcited by 668
  7. 2020Towards Faithfully Interpretable NLP Systems: How should we define and evaluate faithfulness?cited by 590
  8. 2019ERASER: A Benchmark to Evaluate Rationalized NLP Modelscited by 523
  9. 2018e-SNLI: Natural Language Inference with Natural Language Explanationscited by 468
  10. 2019Explain Yourself! Leveraging Language Models for Commonsense Reasoningcited by 367
  11. 2022Selection-Inference: Exploiting Large Language Models for Interpretable Logical Reasoningcited by 331
  12. 2022The Alignment Problem from a Deep Learning Perspectivecited by 246
17 more
  1. 2020Measuring Association Between Labels and Free-Text Rationalescited by 159
  2. 2022Faithful Reasoning Using Large Language Modelscited by 130
  3. 2021Teach Me to Explain: A Review of Datasets for Explainable Natural Language Processingcited by 126
  4. 2022Goal Misgeneralization: Why Correct Specifications Aren't Enough For Correct Goalscited by 120
  5. 2023Faithfulness Tests for Natural Language Explanationscited by 107
  6. 2023Question Decomposition Improves the Faithfulness of Model-Generated Reasoningcited by 107
  7. 2020On Interpretability of Artificial Neural Networks: A Surveycited by 88
  8. 2023On Measuring Faithfulness or Self-consistency of Natural Language Explanationscited by 71
  9. 2023Preventing Language Models From Hiding Their Reasoningcited by 62
  10. 2020Interpretability and Explainability: A Machine Learning Zoo Mini-tourcited by 55
  11. 2020SemEval-2020 Task 4: Commonsense Validation and Explanationcited by 42
  12. 2023Honesty Is the Best Policy: Defining and Mitigating AI Deceptioncited by 40
  13. 2023Challenges with unsupervised LLM knowledge discoverycited by 30
  14. 2022Explaining Chest X-ray Pathologies in Natural Languagecited by 27
  15. 2022Interpretable by Design: Learning Predictors by Composing Interpretable Queriescited by 26
  16. 2020The Struggles of Feature-Based Explanations: Shapley Values vs. Minimal Sufficient Subsetscited by 14
  17. 2022HuSpaCy: an industrial-strength Hungarian natural language processing toolkitcited by 8

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 5; 2025: 13; 2026: 11.

29 papers.

Sort:

By area:

  1. 2025Chain-of-Thought Reasoning In The Wild Is Not Always Faithfulcited by 166
  2. 2024FaithEval: Can Your Language Model Stay Faithful to Context, Even If "The Moon is Made of Marshmallows"cited by 85
  3. 2024A Primer on the Inner Workings of Transformer-based Language Modelscited by 83
  4. 2025Measuring Chain of Thought Faithfulness by Unlearning Reasoning Stepscited by 43
  5. 2025DarkBench: Benchmarking Dark Patterns in Large Language Modelscited by 25
  6. 2025FaithCoT-Bench: Benchmarking Instance-Level Faithfulness of Chain-of-Thought Reasoningcited by 25
  7. 2025How does Chain of Thought Think? Mechanistic Interpretability of Chain-of-Thought Reasoning with Sparse Autoencodingcited by 19
  8. 2025LLMs Can Covertly Sandbag on Capability Evaluations Against Chain-of-Thought Monitoringcited by 17
  9. 2026RAGEN-2: Reasoning Collapse in Agentic RLcited by 16
  10. 2024Towards Faithful Natural Language Explanations: A Study Using Activation Patching in Large Language Modelscited by 14
  11. 2026A Positive Case for Faithfulness: LLM Self-Explanations Help Predict Model Behaviorcited by 12
  12. 2026Biases in the Blind Spot: Detecting What LLMs Fail to Mentioncited by 8
17 more
  1. 2024Fool Me Once? Contrasting Textual and Visual Explanations in a Clinical Decision-Support Settingcited by 5
  2. 2025Towards Transparent Reasoning: What Drives Faithfulness in Large Language Models?cited by 5
  3. 2024How Can We Diagnose and Treat Bias in Large Language Models for Clinical Decision-Making?cited by 4
  4. 2025Aligning What LLMs Do and Say: Towards Self-Consistent Explanationscited by 4
  5. 2026VisualThink-VLA: Visual Intermediate Reasoning for Effective and Low-Latency Vision-Language-Action Policiescited by 4
  6. 2026Argumentation for Explainable and Globally Contestable Decision Support with LLMscited by 3
  7. 2026Faithfulness Metrics Don't Measure Faithfulness: A Meta-Evaluation with Ground Truthcited by 3
  8. 2025Argument-Based Consistency in Toxicity Explanations of LLMscited by 2
  9. 2025Multi-Step Knowledge Interaction Analysis via Rank-2 Subspace Disentanglementcited by 1
  10. 2025Cross-Examiner: Evaluating Consistency of Large Language Model-Generated Explanationscited by 0
  11. 2025NeuroFaith: Evaluating LLM Self-Explanation Faithfulness via Internal Representation Alignmentcited by 0
  12. 2025CAuSE: Decoding Multimodal Classifiers using Faithful Natural Language Explanationcited by 0
  13. 2026WeNLEX: Weakly Supervised Natural Language Explanations for Multilabel Chest X-ray Classificationcited by 0
  14. 2026Faithfulness Serum: Mitigating the Faithfulness Gap in Textual Explanations of LLM Decisions via Attribution Guidancecited by 0
  15. 2026Training Large Language Models for Self-Explanation Faithfulnesscited by 0
  16. 2026Correct Diagnosis, Better Feedback: A Symbolic-Verifier for Faithful LLM Tutoring Feedback in Logic Proofscited by 0
  17. 2026Does a model's stated reason for rejecting a candidate do any work?cited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025A Causal Lens for Evaluating Faithfulness Metrics
  2. 2025A Necessary Step toward Faithfulness: Measuring and Improving Consistency in Free-Text Explanations
  3. 2024Faithfulness vs. Plausibility: On the (Un)Reliability of Explanations from Large Language Models
  4. 2023Benchmarking Faithfulness: Towards Accurate Natural Language Explanations in Vision-Language Tasks
  5. 2026ICE: Intervention-Consistent Explanation Evaluation with Statistical Grounding for LLMs
  6. 2023Do Models Explain Themselves? Counterfactual Simulatability of Natural Language Explanations
  7. 2024Are self-explanations from Large Language Models faithful?
  8. 2025Explanation-Driven Counterfactual Testing for Faithfulness in Vision-Language Model Explanations