mlsanityData to 14 Oct 2026

Mutual Information Alleviates Hallucinations in Abstractive Summarization

arXiv 2210.13210 ↗First posted 24 Oct 2022cs.CLcited by 68

Not on the map. Why?

Abstract (arXiv)

Despite significant progress in the quality of language generated from abstractive summarization models, these models still exhibit the tendency to hallucinate, i.e., output content not supported by the source document. A number of works have tried to fix--or at least uncover the source of--the problem with limited success. In this paper, we identify a simple criterion under which models are significantly more likely to assign more probability to hallucinated content during generation: high model uncertainty. This finding offers a potential explanation for hallucinations: models default to favoring text with high marginal probability, i.e., high-frequency occurrences in the training set, when uncertain about a continuation. It also motivates possible routes for real-time intervention during decoding to prevent such hallucinations. We propose a decoding strategy that switches to optimizing for pointwise mutual information of the source and target token--rather than purely the probability of the target token--when the model exhibits uncertainty. Experiments on the XSum dataset show that our method decreases the probability of hallucinated tokens while maintaining the Rouge and BertS scores of top-performing decoding strategies.

Key facts. Cited by 68 papers in the map (data through 14 Oct 2026). Builds on 31 of its 46 references that are in the map. First posted on arXiv 24 Oct 2022.

Builds on

31 of 46 references.

References · 31

  1. 2020Language Models are Few-Shot Learnerscited by 32,419
  2. 2019BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehensioncited by 5,893
  3. 2019BERTScore: Evaluating Text Generation with BERTcited by 5,512
  4. 2015Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learningcited by 4,448
  5. 2015Neural Machine Translation of Rare Words with Subword Unitscited by 2,918
  6. 2019fairseq: A Fast, Extensible Toolkit for Sequence Modelingcited by 1,488
  7. 2018Don't Give Me the Details, Just the Summary! Topic-Aware Convolutional Neural Networks for Extreme Summarizationcited by 1,317
  8. 2015A Diversity-Promoting Objective Function for Neural Conversation Modelscited by 1,077
  9. 2015Teaching Machines to Read and Comprehendcited by 1,049
  10. 2019Aleatoric and Epistemic Uncertainty in Machine Learning: An Introduction to Concepts and Methodscited by 913
  11. 2020On Faithfulness and Factuality in Abstractive Summarizationcited by 900
  12. 2020SummEval: Re-evaluating Summarization Evaluationcited by 674
19 more
  1. 2019Evaluating the Factual Consistency of Abstractive Text Summarizationcited by 584
  2. 2019Defending Against Neural Fake Newscited by 559
  3. 2021Towards Understanding and Mitigating Social Biases in Language Modelscited by 296
  4. 2021Understanding Factuality in Abstractive Summarization with FRANK: A Benchmark for Factuality Metricscited by 287
  5. 2021All That's 'Human' Is Not Gold: Evaluating Human Evaluation of Generated Textcited by 287
  6. 2021Surface Form Competition: Why the Highest Probability Answer Isn't Always Rightcited by 263
  7. 2018Adaptive Input Representations for Neural Language Modelingcited by 226
  8. 2020Detecting Hallucinated Content in Conditional Neural Sequence Generationcited by 152
  9. 2020On Exposure Bias, Hallucination and Domain Shift in Neural Machine Translationcited by 127
  10. 2020Evaluating Factuality in Generation with Dependency-level Entailmentcited by 122
  11. 2021Improving Faithfulness in Abstractive Summarization with Contrast Candidate Generation and Selectioncited by 76
  12. 2020Improved Natural Language Generation via Loss Truncationcited by 75
  13. 2021A Statistical Analysis of Summarization Evaluation Metrics using Resampling Methodscited by 72
  14. 2020Analyzing the Source and Target Contributions to Predictions in Neural Machine Translationcited by 71
  15. 2021Frequency Effects on Syntactic Rule Learning in Transformerscited by 59
  16. 2020Best-First Beam Searchcited by 40
  17. 2020Understanding Neural Abstractive Summarization Models via Uncertaintycited by 34
  18. 2021Focus Attention: Promoting Faithfulness and Diversity in Summarizationcited by 30
  19. 2020Constrained Abstractive Summarization: Preserving Factual Consistency with Constrained Generationcited by 26

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2022: 2; 2023: 16; 2024: 20; 2025: 20; 2026: 10.

68 papers.

Sort:

By area:

  1. 2023A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questionscited by 1,601
  2. 2023Trusting Your Evidence: Hallucinate Less with Context-aware Decodingcited by 327
  3. 2024Personal LLM Agents: Insights and Survey about the Capability, Efficiency and Securitycited by 260
  4. 2023A Survey of Confidence Estimation and Calibration in Large Language Modelscited by 256
  5. 2024Multi-Modal Hallucination Control by Visual Information Groundingcited by 208
  6. 2024Fact-Checking the Output of Large Language Models via Token-Level Uncertainty Quantificationcited by 162
  7. 2023LM-Polygraph: Uncertainty Estimation for Language Modelscited by 131
  8. 2024Benchmarking Uncertainty Quantification Methods for Large Language Models with LM-Polygraphcited by 105
  9. 2023Enhancing Uncertainty-Based Hallucination Detection with Stronger Focuscited by 99
  10. 2022Evaluating the Factual Consistency of Large Language Models Through News Summarizationcited by 92
  11. 2024The Dawn After the Dark: An Empirical Study on Factuality Hallucination in Large Language Modelscited by 88
  12. 2024Navigating the Risks: A Survey of Security, Privacy, and Ethics Threats in LLM-Based Agentscited by 65
38 more
  1. 2023Uncertainty in Natural Language Generation: From Theory to Applicationscited by 61
  2. 2024AI Hallucinations: A Misnomer Worth Clarifyingcited by 34
  3. 2025Adaptive Retrieval Without Self-Knowledge? Bringing Uncertainty Back Homecited by 33
  4. 2023Faithfulness-Aware Decoding Strategies for Abstractive Summarizationcited by 31
  5. 2024On the Challenges and Opportunities in Generative AIcited by 31
  6. 2025TokUR: Token-Level Uncertainty Estimation for Large Language Model Reasoningcited by 24
  7. 2024On Positional Bias of Faithfulness for Long-form Summarizationcited by 22
  8. 2024Entropy-Based Decoding for Retrieval-Augmented Large Language Modelscited by 19
  9. 2025Uncertainty Quantification for Language Models: A Suite of Black-Box, White-Box, LLM Judge, and Ensemble Scorerscited by 18
  10. 2024An Audit on the Perspectives and Challenges of Hallucinations in NLPcited by 17
  11. 2025Grounding Language with Vision: A Conditional Mutual Information Calibrated Decoding Strategy for Reducing Hallucinations in LVLMscited by 17
  12. 2024Efficient Non-Parametric Uncertainty Quantification for Black-Box Large Language Models and Decision Planningcited by 16
  13. 2023Where's the Liability in Harmful AI Speech?cited by 12
  14. 2023Context-aware Decoding Reduces Hallucination in Query-focused Summarizationcited by 11
  15. 2024REAL Sampling: Boosting Factuality and Diversity of Open-Ended Generation via Asymptotic Entropycited by 11
  16. 2025Uncertainty-Aware Decoding with Minimum Bayes Riskcited by 11
  17. 2025Uncertainty Quantification for Hallucination Detection in Large Language Models: Foundations, Methodology, and Future Directionscited by 11
  18. 2022Improved Beam Search for Hallucination Mitigation in Abstractive Summarizationcited by 10
  19. 2025Read Your Own Mind: Reasoning Helps Surface Self-Confidence Signals in LLMscited by 10
  20. 2025Dynamic Attention-Guided Context Decoding for Mitigating Context Faithfulness Hallucinations in Large Language Modelscited by 9
  21. 2025MAMM-Refine: A Recipe for Improving Faithfulness in Generation with Multi-Agent Collaborationcited by 9
  22. 2024HELPD: Mitigating Hallucination of LVLMs by Hierarchical Feedback Learning with Vision-enhanced Penalty Decodingcited by 8
  23. 2023Neural Natural Language Processing for Long Texts: A Survey on Classification and Summarizationcited by 7
  24. 2023Hallucination Detection for Grounded Instruction Generationcited by 7
  25. 2023Adapter-based Selective Knowledge Distillation for Federated Multi-domain Meeting Summarizationcited by 6
  26. 2024textttCOSMIC: Mutual Information for Task-Agnostic Summarization Evaluationcited by 6
  27. 2024Mitigating Hallucination in Abstractive Summarization with Domain-Conditional Mutual Informationcited by 6
  28. 2024Estimating Privacy Leakage of Augmented Contextual Knowledge in Language Modelscited by 6
  29. 2024Diver: Large Language Model Decoding with Span-Level Mutual Information Verificationcited by 5
  30. 2025SCOPE: A Self-supervised Framework for Improving Faithfulness in Conditional Text Generationcited by 5
  31. 2025Probabilities Are All You Need: A Probability-Only Approach to Uncertainty Estimation in Large Language Modelscited by 5
  32. 2023Generating Faithful Text From a Knowledge Graph with Noisy Reference Textcited by 4
  33. 2026LLMs Exhibit Significantly Lower Uncertainty in Creative Writing Than Professional Writerscited by 4
  34. 2025A Rate-Distortion Framework for Summarizationcited by 3
  35. 2025Gradient-guided Attention Map Editing: Towards Efficient Contextual Hallucination Mitigationcited by 3
  36. 2026Semantic Calibration Prevails Where Token Confidence Fails: Benchmarking Long-Form Scientific QAcited by 3
  37. 2023Anti-LM Decoding for Zero-shot In-context Machine Translationcited by 2
  38. 2024On the Benefits of Fine-Grained Loss Truncation: A Case Study on Factuality in Summarizationcited by 2

50 most cited of 68 All 68 (JSON)