mlsanityData to 14 Oct 2026

Distributional Semantics Tracing: A Framework for Explaining Hallucinations in Large Language Models

arXiv 2510.06107 ↗First posted 7 Oct 2025cs.CL cs.AI cs.CEcited by 3

Not on the map. Why?

Abstract (arXiv)

Hallucinations in large language models (LLMs) produce fluent continuations that are not supported by the prompt, especially under minimal contextual cues and ambiguity. We introduce Distributional Semantics Tracing (DST), a model-native method that builds layer-wise semantic maps at the answer position by decoding residual-stream states through the unembedding, selecting a compact top-K concept set, and estimating directed concept-to-concept support via lightweight causal tracing. Using these traces, we test a representation-level hypothesis: hallucinations arise from correlation-driven representational drift across depth, where the residual stream is pulled toward a locally coherent but context-inconsistent concept neighborhood reinforced by training co-occurrences. On Racing Thoughts dataset, DST yields more faithful explanations than attribution, probing, and intervention baselines under an LLM-judge protocol, and the resulting Contextual Alignment Score (CAS) strongly predicts failures, supporting this drift hypothesis.

Key facts. Cited by 3 papers in the map (data through 14 Oct 2026). Builds on 71 of its 112 references that are in the map. First posted on arXiv 7 Oct 2025.

Builds on

71 of 112 references.

References · 71

  1. 2017Attention Is All You Needcited by 60,361
  2. 2017A Unified Approach to Interpreting Model Predictionscited by 6,555
  3. 2016"Why Should I Trust You?": Explaining the Predictions of Any Classifiercited by 5,070
  4. 2017Axiomatic Attribution for Deep Networkscited by 3,384
  5. 2021Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsitycited by 3,073
  6. 2022Locating and Editing Factual Associations in GPTcited by 2,691
  7. 2018Deep contextualized word representationscited by 2,290
  8. 2023Efficient Streaming Language Models with Attention Sinkscited by 2,186
  9. 2020Shortcut Learning in Deep Neural Networkscited by 1,969
  10. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,420
  11. 2020Transformer Feed-Forward Layers Are Key-Value Memoriescited by 1,336
  12. 2023FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generationcited by 1,282
38 more
  1. 2022Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 smallcited by 1,165
  2. 2024The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scalecited by 1,160
  3. 2019Right for the Wrong Reasons: Diagnosing Syntactic Heuristics in Natural Language Inferencecited by 876
  4. 2023The RefinedWeb Dataset for Falcon LLM: Outperforming Curated Corpora with Web Data, and Web Data Onlycited by 763
  5. 2023Eliciting Latent Predictions from Transformers with the Tuned Lenscited by 623
  6. 2019ERASER: A Benchmark to Evaluate Rationalized NLP Modelscited by 530
  7. 2023Model Tells You What to Discard: Adaptive KV Cache Compression for LLMscited by 474
  8. 2023Scissorhands: Exploiting the Persistence of Importance Hypothesis for LLM KV Cache Compression at Test Timecited by 439
  9. 2024Mechanistic Interpretability for AI Safety -- A Reviewcited by 422
  10. 2023How Language Model Hallucinations Can Snowballcited by 311
  11. 2024LayerSkip: Enabling Early Exit Inference and Self-Speculative Decodingcited by 307
  12. 2024Do Llamas Work in English? On the Latent Language of Multilingual Transformerscited by 281
  13. 2024A Comprehensive Survey of Hallucination Mitigation Techniques in Large Language Modelscited by 269
  14. 2024LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinationscited by 244
  15. 2025Reasoning with Latent Thoughts: On the Power of Looped Transformerscited by 240
  16. 2024Patchscopes: A Unifying Framework for Inspecting Hidden Representations of Language Modelscited by 235
  17. 2019Probing Neural Network Comprehension of Natural Language Argumentscited by 214
  18. 2022Factuality Enhanced Language Models for Open-Ended Text Generationcited by 202
  19. 2025Reasoning Models Know When They're Right: Probing Hidden States for Self-Verificationcited by 159
  20. 2023Investigating the Factual Knowledge Boundary of Large Language Models with Retrieval Augmentationcited by 147
  21. 2024Devils in Middle Layers of Large Vision-Language Models: Interpreting, Detecting and Mitigating Object Hallucinations via Attention Lenscited by 143
  22. 2022On the Origin of Hallucinations in Conversational Models: Is it the Datasets or the Models?cited by 138
  23. 2020An Empirical Study on Robustness to Spurious Correlations using Pre-trained Language Modelscited by 134
  24. 2021Entity-level Factual Consistency of Abstractive Text Summarizationcited by 118
  25. 2019Assessing The Factual Accuracy of Generated Textcited by 103
  26. 2024LLM Internal States Reveal Hallucination Risk Faced With a Querycited by 82
  27. 2024Self-Alignment for Factuality: Mitigating Hallucinations in LLMs via Self-Evaluationcited by 80
  28. 2025Do Multilingual LLMs Think In English?cited by 78
  29. 2023Large Language Models Can be Lazy Learners: Analyze Shortcuts in In-Context Learningcited by 69
  30. 2022Understanding Transformer Memorization Recall Through Idiomscited by 62
  31. 2024Demystifying Verbatim Memorization in Large Language Modelscited by 56
  32. 2024Usable XAI: 10 Strategies Towards Exploiting Explainability in the LLM Eracited by 46
  33. 2025Hallucination Detection in LLMs Using Spectral Features of Attention Mapscited by 43
  34. 2024Summing Up the Facts: Additive Mechanisms Behind Factual Recall in LLMscited by 42
  35. 2021Why Machine Reading Comprehension Models Learn Shortcuts?cited by 40
  36. 2024Mechanistic Understanding and Mitigation of Language Model Non-Factual Hallucinationscited by 38
  37. 2025HALoGEN: Fantastic LLM Hallucinations and Where to Find Themcited by 34
  38. 2025Everything, Everywhere, All at Once: Is Mechanistic Interpretability Identifiable?cited by 34

50 most cited of 71 All 71 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 1; 2026: 2.

3 papers.

  1. 2025A Unified Definition of Hallucination: It's The World Model, Stupid!cited by 6
  2. 2026Detecting Hallucinations in Retrieval-Augmented Generation via Semantic-level Internal Reasoning Graphcited by 2
  3. 2026HART: Data-Driven Hallucination Attribution and Evidence-Based Tracing for Large Language Modelscited by 0