mlsanityData to 7 Oct 2026

Strong hallucinations from negation and how to fix them

arXiv 2402.10543 ↗First posted 16 Feb 2024cs.CL cs.AIcited by 6

On the map: LLM reasoning → Hallucination detection and mitigation

Abstract (arXiv)

Despite great performance on many tasks, language models (LMs) still struggle with reasoning, sometimes providing responses that cannot possibly be true because they stem from logical incoherence. We call such responses textitstrong hallucinations and prove that they follow from an LM's computation of its internal representations for logical operators and outputs from those representations. Focusing on negation, we provide a novel solution in which negation is treated not as another element of a latent representation, but as textitan operation over an LM's latent representations that constrains how they may evolve. We show that our approach improves model performance in cloze prompting and natural language inference tasks with negation without requiring training on sparse negative data.

Key facts. Cited by 6 papers in the map (data through 7 Oct 2026). Placed in Hallucination detection and mitigation, one of 17 directions in LLM reasoning. Builds on 39 of its 92 references that are in the map. First posted on arXiv 16 Feb 2024.

Builds on

39 of 92 references.

References · 39

  1. 2017Attention Is All You Needcited by 59,932
  2. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  3. 2019RoBERTa: A Robustly Optimized BERT Pretraining Approachcited by 13,266
  4. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,916
  5. 2019ViLBERT: Pretraining Task-Agnostic Visiolinguistic Representations for Vision-and-Language Taskscited by 2,219
  6. 2023PaLM-E: An Embodied Multimodal Language Modelcited by 2,037
  7. 2019Language Models as Knowledge Bases?cited by 1,922
  8. 2015A large annotated corpus for learning natural language inferencecited by 1,837
  9. 2021WebGPT: Browser-assisted question-answering with human feedbackcited by 1,722
  10. 2019VisualBERT: A Simple and Performant Baseline for Vision and Languagecited by 1,291
  11. 2020BLEURT: Learning Robust Metrics for Text Generationcited by 1,196
  12. 2021Cross-Task Generalization via Natural Language Crowdsourcing Instructionscited by 696
27 more
  1. 2018CoQA: A Conversational Question Answering Challengecited by 674
  2. 2019What do you learn from context? Probing for sentence structure in contextualized word representationscited by 436
  3. 2021Entity-Based Knowledge Conflicts in Question Answeringcited by 349
  4. 2020ToTTo: A Controlled Table-To-Text Generation Datasetcited by 267
  5. 2019Negated and Misprimed Probes for Pretrained Language Models: Birds Can Talk, But Cannot Flycited by 240
  6. 2021Implicit Representations of Meaning in Neural Language Modelscited by 179
  7. 2019oLMpics -- On what Language Model Pre-training Capturescited by 161
  8. 2023LINC: A Neurosymbolic Approach for Logical Reasoning by Combining Language Models with First-Order Logic Proverscited by 161
  9. 2017Faithful to the Original: Fact Aware Neural Abstractive Summarizationcited by 141
  10. 2020Evaluating Factuality in Generation with Dependency-level Entailmentcited by 122
  11. 2019Handling Divergent Reference Texts when Evaluating Table-to-Text Generationcited by 119
  12. 2020Logical Neural Networkscited by 118
  13. 2020Controlled Hallucinations: Learning to Generate Faithfully from Noisy Datacited by 83
  14. 2021The Factual Inconsistency Problem in Abstractive Text Summarization: A Surveycited by 79
  15. 2021Understanding by Understanding Not: Modeling Negation in Language Modelscited by 73
  16. 2023Language models are not naysayers: An analysis of language models on negation benchmarkscited by 73
  17. 2020Knowledge Graph-Augmented Abstractive Summarization with Semantic-Driven Cloze Rewardcited by 66
  18. 2023Tighter Bounds on the Expressivity of Transformer Encoderscited by 65
  19. 2020From Statistical Relational to Neuro-Symbolic Artificial Intelligencecited by 63
  20. 2020Deep Learning and the Global Workspace Theorycited by 35
  21. 2022Human Evaluation and Correlation with Automatic Metrics in Consultation Note Generationcited by 33
  22. 2020An information theoretic view on selecting linguistic probescited by 19
  23. 2020Braid: Weaving Symbolic and Neural Knowledge into Coherent Logical Explanationscited by 11
  24. 2023Semi-supervised Multimodal Representation Learning through a Global Workspacecited by 11
  25. 2022Beyond Distributional Hypothesis: Let Language Models Learn Meaning-Text Correspondencecited by 10
  26. 2022Entailment Semantics Can Be Extracted from an Ideal Language Modelcited by 10
  27. 2022Analyzing Semantic Faithfulness of Language Models via Input Intervention on Question Answeringcited by 6

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 4; 2026: 2.

6 papers.

  1. 2025Hallucination Detection in Large Language Models with Metamorphic Relationscited by 24
  2. 2025Investigating Symbolic Triggers of Hallucination in Gemma Models Across HaluEval and TruthfulQAcited by 3
  3. 2025On the Thinking-Language Modeling Gap in Large Language Modelscited by 0
  4. 2025The Impact of Negated Text on Hallucination with Large Language Modelscited by 0
  5. 2026Walking on the DARKSIDEcited by 0
  6. 2026Understanding as No-Arbitrage: Bounded Dutch Books as a Definition and Training Objective for Language Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024Investigating and Addressing Hallucinations of LLMs in Tasks Involving Negation
  2. 2025Negation: A Pink Elephant in the Large Language Models' Room?
  3. 2026Listen to the Layers: Mitigating Hallucinations with Inter-Layer Disagreement
  4. 2025The Hallucination Tax of Reinforcement Finetuning
  5. 2026Why LLMs Hallucinate on Structured Knowledge: A Mechanistic Analysis of Reasoning over Linearized Representations
  6. 2024Benchmarking Hallucination in Large Language Models based on Unanswerable Math Word Problem
  7. 2022Discovering Latent Knowledge in Language Models Without Supervision
  8. 2026Score × Decoder: A Unified View of Unsupervised Inference-Time Scaling for Hallucination Mitigation