Pith. sign in

REVIEW 3 major objections 5 minor 41 references

Exploring Causal Effect of Social Bias on Faithfulness Hallucinations in Large Language Models

T0 review · 3 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Social bias causally shifts whether LLM outputs stay faithful to the input, with anti-stereotype phrasing increasing errors.

desk verdict Useful dataset and a clean empirical trend, but the causal claim outruns the intervention: swapping attribute words isn't a validated do-operation. read the letter →

arxiv 2508.07753 v1 pith:D63N7L2S submitted 2025-08-11 cs.CL

Zhenliang Zhang , Junzhe Zhang , Xinyu Hu , HuiXuan Zhang , Xiaojun Wan This is my paper · ORCID
classification cs.CL
keywords faithfulnesshallucinationsocialbiascausalinferencestructuralmodelinterventionunfairnesslargelanguagemodelsstatisticalsignificancetesting
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that the social-bias framing of a text—whether it matches, contradicts, or avoids a stereotype—is a cause, not merely a correlate, of faithfulness hallucinations in large language models. The authors build a structural causal model and design controlled bias interventions that rewrite the same scenario into pro-stereotype, anti-stereotype, and non-stereotype versions differing only in social attributes. Across seven models and five bias categories, they find that anti-stereotype wording increases hallucinations, pro-stereotype wording suppresses them, and the effect is concentrated in a newly named class, unfairness hallucinations. If right, this means bias is not a tangential fairness concern but a direct driver of model unfaithfulness, and it can hide inside confident, hard-to-detect errors.

What carries the argument

The central machinery is a three-state bias intervention built on a structural causal model: each scenario has a bias-state node $B$, a hallucination node $H$, and a confounder node $Z$, and the intervention $\mathrm{do}(B=\cdot)$ sets the bias state directly while cutting paths from confounders. Paired templates make the intervention concrete by rewriting the same scene into Pro-, Anti-, and Non-stereotype versions that differ only in the social attributes of the characters. The Individual Causal Effect (ICE) records whether the binary hallucination outcome changes between two versions; summing ICEs and running McNemar's test yields the Unified Causal Significance (UCS), which reports both

What would settle it

Take the Pro-Anti pairs from the dataset and replace the social-attribute words with alternative synonyms matched for token frequency, length, and association strength. If the hallucination-rate gap between pro- and anti-stereotype versions disappears or reverses under frequency-matched edits, the intervention is not isolating bias state. A simpler companion check: regress hallucination outcome on bias state and on the token log-frequency of the swapped attribute words; if token frequency absorbs the bias-state effect, the causal claim fails.

Watch

Extended reading notes

Core claim

The paper's central claim is that bias states directly cause faithfulness hallucinations in LLMs, not just correlate with them. Using paired texts that differ only in social attributes, it reports that shifting from non-stereotype to anti-stereotype raises hallucination rates significantly in 34 of 35 model-bias combinations, shifting from pro-stereotype to anti-stereotype raises them in 32 of 35, while pro-stereotype phrasing suppresses hallucinations relative to non-stereotype in 19 of 35 combinations. The same significance analysis shows the causal effect falls almost entirely on unfairness hallucinations—errors where the model chooses a person whose social attribute is inconsistent with

Load-bearing premise

The load-bearing premise is that swapping only the social attributes in a paired text leaves every other feature—topic, sentiment, word frequency, syntax—unchanged; the paper asserts this precision but never measures it directly, so if the swaps change other things, the causal difference is really a correlation with the whole text edit.

Editorial extensions

If this is right

  • Bias-induced hallucination is a distinct failure mode: two models with similar hallucination rates can have very different causal sensitivity to bias, so aggregate accuracy is not a reliable proxy.
  • Deployment of QA systems should treat anti-stereotype contexts as higher-risk inputs for unfaithful answers.
  • Evaluation sets built from stereotype-aligned text can overstate faithfulness because pro-stereotype prompts suppress errors.
  • Unfairness hallucinations deserve separate reporting in hallucination benchmarks; confidence-based filters will miss them.
  • The same intervention-and-test recipe can be applied to other suspected causes of context-driven hallucinations, such as information position or irrelevant content.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The causal claim depends on the precision of the intervention, which the paper asserts but does not directly measure. In the worked examples the swapped words are near-synonyms with different lexical frequencies, so part of the effect could come from token-level confounds rather than bias state itself.
  • A direct test of that concern would be to repeat the comparisons using attribute words matched on frequency and surprisal; if the Pro-Anti gap persists, the causal story is strengthened, and if it shrinks, the bias-state effect is partly lexical.
  • The consistent direction of the effect suggests a mechanism: stereotype-consistent text aligns with the model's prior and helps it stay on task, while stereotype-contradicting text fights that prior and increases error. This could be probed by tracking internal representations across the three bias states.
  • The framework could rank other context features by causal significance, producing an ordered map of what makes LLMs unfaithful rather than a single correlation.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper claims that social bias states (anti-stereotype, pro-stereotype, non-stereotype) have a causal effect on faithfulness hallucinations in large language models. It formalizes the problem with a Structural Causal Model, defines bias interventions via template-based text edits, and proposes ICE/UCS statistics tested with McNemar's test. The authors construct a new dataset, BID, from BBQ scenarios and evaluate seven LLMs. They report that anti-stereotype contexts increase hallucinations while pro-stereotype contexts suppress them, and they introduce a new category called 'unfairness hallucination'. The central claim is that this is the first direct causal demonstration, going beyond correlation.

Significance. If the causal claim is established, the paper would be a valuable contribution: it takes a step from correlational studies of hallucination triggers toward interventional analysis, contributes a medium-scale bias intervention dataset, and uses deterministic greedy decoding and paired McNemar tests, which are appropriate for the binary outcome setting. The authors also clearly list the ideal conditions (effectiveness, precision, consistency) that an intervention should satisfy. However, the experiment as reported does not validate those conditions, and the central causal conclusion is currently not supported by the evidence presented.

major comments (3)
  1. [§3.2, Figure 7] The intervention do(B=...) is implemented as lexical substitution of social-attribute words. The Precision criterion in §3.2 requires that all contextual elements except B remain unchanged, but no validation is provided. In Figure 7's Pro-Anti pair, the text changes from 'wealthy/poor' to 'low-income/rich'; these tokens differ in frequency, register, and sentiment, which §3.1 itself lists as hallucination confounders. Since no manipulation check, placebo condition, or covariate analysis is reported, the observed Pro-Anti difference cannot be distinguished from an effect of surface-form perturbation. This undermines Eq. (1) and the central causal claim in §5.2.2.
  2. [§3.2, Figure 7] Consistency is not met for the main Pro-Anti comparison. In Figure 7, Non-Pro and Non-Anti pairs change one attribute token, while the Pro-Anti pair changes two attribute tokens. Thus the Pro-Anti ICE may be inflated by larger textual perturbation, and the directional conclusion 'Anti-stereotype increases, Pro-stereotype suppresses' in §5.2.2 is not identified. The authors should either construct all pairs with identical edit distance and matched token-frequency distributions or conduct a control analysis with 'sham' attribute swaps.
  3. [§5.2.2, Table 3] The language 'Social bias is a key cause of faithfulness hallucinations' overstates what the experiment establishes. Without verification of the intervention criteria, Table 3 reports associations between template edits and model errors, not causal effects. The authors should add validation experiments (e.g., randomized attribute shuffles, token-frequency-matched controls, and a pre-registered analysis plan) and soften causal claims accordingly.
minor comments (5)
  1. [§3.1] Typo: 'we first defines the key concepts' should read 'we first define...'.
  2. [Appendix B.4] The description of one-tailed tests is nonstandard: a chi-square test has no left/right tail as described. Direction should be read from the sign of ∑ICE after a significant two-sided test; please clarify the statistical language.
  3. [§5.1] The text says options are randomly shuffled and multiple generations are performed, but greedy decoding is deterministic. Clarify how randomness affects the reported rates and whether confidence is averaged over option orders.
  4. [Figure 4] The reported ordering Pro < Non < Anti is presented as a trend without error bars or variance estimates. Since greedy decoding is deterministic, repeated sampling from option shuffles could provide a check of stability.
  5. [References / Figure 7] References [24] and [25] are the same paper; [26] appears irrelevant to the word-frequency confounder claim. Also, the Figure 7 caption says the pairs 'differ only in social attributes,' but the lexical realizations differ (e.g., wealthy vs. rich), so the caption overstates the intervention's precision.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the causal claim is empirically tested on independently labeled outputs; the main weakness is an unverified intervention validity assumption, not a circular reduction.

full rationale

The paper's derivation chain is not circular. Bias states (Pro/Anti/Non) are defined a priori from attested social stereotypes and template attribute assignments (Section 3.1, Figure 1), while the hallucination outcome H is independently measured by whether the model's answer matches the template's keyed correct answer (Section 4, Figure 3). ICE and UCS (Eqs. 1-4) are computed from these independently obtained outputs over paired template variants. No parameter is fitted to hallucination data and then renamed a prediction; BID is manually constructed from BBQ scenarios, not derived from model outputs. The only self-citation ([34], related work on an ICR probe) is used to note that modules correlate with hallucinations and is not load-bearing for the causal claim. The principal weakness is that the 'Precision' and 'Consistency' criteria for the do-intervention are asserted rather than empirically verified—e.g., Figure 7 shows Pro-Anti pairs differing in two attribute tokens ('wealthy/poor' vs 'low-income/rich') while Non-Pro/Non-Anti pairs differ in one, and lexical substitutions change word frequency, which Section 3.1 lists as a confounder. This threatens causal identification but is not a circular reduction: the paper does not define bias state in terms of hallucination outcome, nor does it fit a parameter and call it a prediction. The causal claim could be confounded, but it is not true by construction.

Assumptions & free parameters 1 free parameters · 5 assumptions · 1 invented entities

The central claim rests on the manual design of paired texts and the assumption that the intervention cleanly isolates bias state from other context features. The attribute lexicons and the Unfairness hallucination category are author-chosen constructs that influence measurement but are not fitted to outcome data.

free parameters (1)
  • Social attribute lexicon per bias type
    Each bias state is instantiated by manually chosen attribute words (e.g., high-SES: scientist, doctor; low-SES: bartender, cleaner). The specific choice is unvalidated and may carry lexical confounds (frequency, sentiment, typicality) that affect hallucination rates independently of bias state.
assumptions (5)
  • domain assumption The bias intervention satisfies the Effectiveness, Precision, and Consistency criteria as defined in Section 3.2, so that do(B=...) on templated text isolates the causal effect of B on H.
    This is the load-bearing premise of the causal claim. It is asserted but not empirically validated; there are no confounder checks or equivalence evaluations.
  • domain assumption The three bias states (Pro, Anti, Non) are correctly operationalized by the constructed templates and attribute lexicons, with Non-stereotype genuinely bias-neutral.
    The Non-stereotype condition uses synonymous high- or low-attribute pairs (e.g., 'wealthy/rich', 'low-income/poor'), which may not be truly balanced in model associations.
  • domain assumption Selecting any incorrect option in the forced-choice QA is a valid operationalization of faithfulness hallucination.
    The paper equates hallucination with picking the wrong person in a four-option QA; this ignores cases where the model may produce a correct alternative or where the template itself is ambiguous.
  • standard math McNemar's test assumptions hold: paired binary outcomes and a sufficiently large sample for the chi-square approximation.
    The test is standard, but the approximation can be inaccurate for small b+c; the paper does not report the number of discordant pairs.
  • standard math do-calculus rules apply to the defined causal graph with Z confounding B and H.
    The graph is a standard confounded structure, but the intervention is not a real manipulation of B in the model; it is a text-level proxy, so the mapping from do(B) to template edits is an extra assumption.
invented entities (1)
  • Unfairness hallucination independent evidence
    purpose: A new subtype of faithfulness hallucination where the model selects an individual whose social attribute is unfairly different from the correct answer (e.g., choosing a male when a female is correct in a biased context). Used to scope the causal effect and claim a new phenomenon.
    The category is operationally defined from model outputs and template labels, so it is observable. However, it is a relabeling of existing errors rather than a new theoretical entity, and its independence from the bias-state definition is not fully established since it is only defined in Pro/Anti contexts.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Exploring Causal Effect of Social Bias on Faithfulness Hallucinations in Large Language Models." pith.science (2026). https://pith.science/paper/D63N7L2S

@misc{pith2026250807753,
  author       = {Pith},
  title        = {Pith review of: Exploring Causal Effect of Social Bias on Faithfulness Hallucinations in Large Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/D63N7L2S}},
  note         = {Machine review of arXiv:2508.07753}
}
read the original abstract

Large language models (LLMs) have achieved remarkable success in various tasks, yet they remain vulnerable to faithfulness hallucinations, where the output does not align with the input. In this study, we investigate whether social bias contributes to these hallucinations, a causal relationship that has not been explored. A key challenge is controlling confounders within the context, which complicates the isolation of causality between bias states and hallucinations. To address this, we utilize the Structural Causal Model (SCM) to establish and validate the causality and design bias interventions to control confounders. In addition, we develop the Bias Intervention Dataset (BID), which includes various social biases, enabling precise measurement of causal effects. Experiments on mainstream LLMs reveal that biases are significant causes of faithfulness hallucinations, and the effect of each bias state differs in direction. We further analyze the scope of these causal effects across various models, specifically focusing on unfairness hallucinations, which are primarily targeted by social bias, revealing the subtle yet significant causal effect of bias on hallucination generation.

Figures

Figures reproduced from arXiv: 2508.07753 by the authors.

Figure 1
Figure 1. Illustration of the three bias states. Pro-stereotype [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Left: The original causal graph, where directed edges [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Left Overview of data construction: Templates are designed to include at least three individuals, with two having configurable social attributes and one without. Bias intervention: Social attributes are combined and assigned specific SES values to create contexts with various bias states, ensuring consistency across interventions. Right Pairwise comparison to calculate ICE: Comparing two different bias state context… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Hallucination rates on BID. This figure illustrates the hallucination rates of each model across different bias states: [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Scope of causal effect. UCS between two types of hallucinations (unfairness, common) and bias states, with the red dashed line indicating the significance threshold. The figure shows a significant causal relationship between un￾fairness hallucinations and social bias i…
Figure 6
Figure 6. Figure 6: Average confidence of the LLMs for three types of [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Pairwise comparison, each data pair consists of two texts with different bias states, differing only in the social [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 16 canonical work pages

  1. [1]

    AI@Meta. 2024. Llama 3 Model Card. (2024). https://github.com/meta-llama/ llama3/blob/main/MODEL_CARD.md

    2024

  2. [2]

    Zechen Bai, Pichao Wang, Tianjun Xiao, Tong He, Zongbo Han, Zheng Zhang, and Mike Zheng Shou. 2024. Hallucination of Multimodal Large Language Models: A Survey. ArXiv abs/2404.18930 (2024). https://api.semanticscholar.org/ CorpusID:269449935

    arXiv 2024

  3. [3]

    Adam Bouyamourn. 2023. Why LLMs Hallucinate, and How to Get (Evidential) Closure: Perceptual, Intensional, and Extensional Learning for Faithful Natural Language Generation. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing , Houda Bouamor, Juan Pino, and Kalika Bali (Eds.). Association for Computational Linguistics...

    work page 2023

  4. [4]

    Zhibo Chu, Zichong Wang, and Wenbin Zhang. 2024. Fairness in Large Language Models: A Taxonomic Survey. SIGKDD Explor. Newsl. 26, 1 (jul 2024), 34–48. doi:10.1145/3682112.3682117

    arXiv 2024

  5. [5]

    Yifei Da, Matías Nicolás Bossa, Abel Díaz Berenguer, and Hichem Sahli. 2024. Reducing Bias in Sentiment Analysis Models Through Causal Mediation Analysis and Targeted Counterfactual Training. IEEE Access 12 (2024), 10120–10134. doi:10.1109/ACCESS.2024.3353056

    work page arXiv 2024

  6. [6]

    Guoliang Dong, Haoyu Wang, Jun Sun, and Xinyu Wang. 2024. Evaluating and Mitigating Linguistic Discrimination in Large Language Models. ArXiv abs/2404.18534 (2024). https://api.semanticscholar.org/CorpusID:269449779

    arXiv 2024

  7. [7]

    Minda Hu, Bowei He, Yufei Wang, Liangyou Li, Chen Ma, and Irwin King

    work page

  8. [8]

    Lei Huang, Weijiang Yu, Weitao Ma, Weihong Zhong, Zhangyin Feng, Haotian Wang, Qianglong Chen, Weihua Peng, Xiaocheng Feng, Bing Qin, and Ting Liu. 2023. A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions. arXiv:2311.05232 [cs.CL] https: //arxiv.org/abs/2311.05232

    arXiv 2023

Show all 41 references
  1. [9]

    Ziwei Ji, Nayeon Lee, Rita Frieske, Tiezheng Yu, Dan Su, Yan Xu, Etsuko Ishii, Yejin Bang, Delong Chen, Wenliang Dai, Andrea Madotto, and Pascale Fung

  2. [10]

    Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, De- vendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, Lélio Renard Lavaud, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thoma...

    2023 arXiv

  3. [11]

    Hadas Kotek, Rikker Dockum, and David Sun. 2023. Gender bias and stereotypes in Large Language Models. In Proceedings of The ACM Collective Intelligence Conference (Delft, Netherlands) (CI ’23). Association for Computing Machinery, New York, NY, USA, 12–24. doi:10.1145/3582269.3615599

    2023

  4. [12]

    Faisal Ladhak, Esin Durmus, Mirac Suzgun, Tianyi Zhang, Dan Jurafsky, Kath- leen McKeown, and Tatsunori Hashimoto. 2023. When Do Pre-Training Bi- ases Propagate to Downstream Tasks? A Case Study in Text Summarization. In Proceedings of the 17th Conference of the European Chapt...

    2023

  5. [13]

    He Li, Haoang Chi, Mingyu Liu, and Wenjing Yang. 2024. Look Within, Why LLMs Hallucinate: A Causal Perspective. arXiv:2407.10153 [cs.CL] https://arxiv. org/abs/2407.10153

    2024 arXiv

  6. [14]

    Junyi Li, Xiaoxue Cheng, Xin Zhao, Jian-Yun Nie, and Ji-Rong Wen. 2023. HaluE- val: A Large-Scale Hallucination Evaluation Benchmark for Large Language Models. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, Houda Bouamor, Juan Pino, ...

    2023 doi

  7. [15]

    Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang

    Nelson F. Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang. 2024. Lost in the Middle: How Language Models Use Long Contexts. Transactions of the Association for Computational Linguistics 12 (2024), 157–173. doi:10.1162/tacl_a_00638

    2024 doi

  8. [16]

    Nick McKenna, Tianyi Li, Liang Cheng, Mohammad Hosseini, Mark Johnson, and Mark Steedman. 2023. Sources of Hallucination by Large Language Models on Inference Tasks. In Findings of the Association for Computational Linguistics: EMNLP 2023, Houda Bouamor, Juan Pino, and Kalika ...

    2023 doi

  9. [17]

    Quinn McNemar. 1947. Note on the Sampling Error of the Difference between Correlated Proportions or Percentages. Psychometrika 12, 2 (June 1947), 153–157. doi:10.1007/BF02295996

    1947 doi

  10. [18]

    OpenAI. 2023. GPT-3.5 Turbo: Fine-Tuning and API Updates. https://openai. com/index/gpt-3-5-turbo-fine-tuning-and-api-updates/

    2023

  11. [19]

    OpenAI. 2024. GPT-4o Mini: Advancing Cost-Efficient Intelligence. https:// openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence/

    2024

  12. [20]

    Alicia Parrish, Angelica Chen, Nikita Nangia, Vishakh Padmakumar, Jason Phang, Jana Thompson, Phu Mon Htut, and Samuel Bowman. 2022. BBQ: A hand- built bias benchmark for question answering. In Findings of the Association for Computational Linguistics: ACL 2022 , Smaranda Mure...

    2022 doi

  13. [21]

    Judea Pearl. 2010. An introduction to causal inference. The international journal of biostatistics 6, 2 (2010)

    2010

  14. [22]

    Chahat Raj, Anjishnu Mukherjee, Aylin Caliskan, Antonios Anastasopoulos, and Ziwei Zhu. 2024. Breaking Bias, Building Bridges: Evaluation and Mitigation of Social Biases in LLMs via Contact Hypothesis. AAAI/ACM conference on AI, Ethics, and Society (2024)

    2024

  15. [23]

    Beatrice Savoldi, Marco Gaido, Luisa Bentivogli, Matteo Negri, and Marco Turchi

  16. [24]

    Freda Shi, Xinyun Chen, Kanishka Misra, Nathan Scales, David Dohan, Ed Chi, Nathanael Schärli, and Denny Zhou. 2023. Large language models can be easily distracted by irrelevant context. InProceedings of the 40th International Conference on Machine Learning (Honolulu, Hawaii, ...

    2023

  17. [25]

    Freda Shi, Xinyun Chen, Kanishka Misra, Nathan Scales, David Dohan, Ed H Chi, Nathanael Schärli, and Denny Zhou. 2023. Large language models can be easily distracted by irrelevant context. In International Conference on Machine Learning. PMLR, 31210–31227

    2023

  18. [26]

    An Tang, Xiuzhen Zhang, and Minh Dinh. 2024. Aspect-based Key Point Analysis for Quantitative Summarization of Reviews. In Findings of the Association for Computational Linguistics: EACL 2024, Yvette Graham and Matthew Purver (Eds.). Association for Computational Linguistics, ...

    2024

  19. [27]

    Gemma Team, Morgane Riviere, Shreya Pathak, Pier Giuseppe Sessa, Cassidy Hardin, Surya Bhupatiraju, Léonard Hussenot, Thomas Mesnard, Bobak Shahri- ari, Alexandre Ramé, Johan Ferret, Peter Liu, Pouya Tafti, Abe Friesen, Michelle Zhang et al. Casbon, Sabela Ramos, Ravin Kumar, ...

    2024 arXiv

  20. [28]

    Qwen Team. 2024. Qwen2.5: A Party of Foundation Models. https://qwenlm. github.io/blog/qwen2.5/

    2024

  21. [29]

    Kelly is a Warm Person, Joseph is a Role Model

    Yixin Wan, George Pu, Jiao Sun, Aparna Garimella, Kai-Wei Chang, and Nanyun Peng. 2023. “Kelly is a Warm Person, Joseph is a Role Model”: Gender Biases in LLM-Generated Reference Letters. In Findings of the Association for Com- putational Linguistics: EMNLP 2023 , Houda Bouamo...

    2023 doi

  22. [30]

    Kankanhalli

    Ziwei Xu, Sanjay Jain, and Mohan S. Kankanhalli. 2024. Hallucination is In- evitable: An Innate Limitation of Large Language Models. ArXiv abs/2401.11817 (2024). https://api.semanticscholar.org/CorpusID:267069207

    2024 arXiv

  23. [31]

    Muru Zhang, Ofir Press, William Merrill, Alisa Liu, and Noah A Smith. 2023. How language model hallucinations can snowball. arXiv preprint arXiv:2305.13534 (2023)

    2023 arXiv

  24. [32]

    Fung, Jing Li, Manling Li, and Heng Ji

    Yuji Zhang, Sha Li, Jiateng Liu, Pengfei Yu, Yi R. Fung, Jing Li, Manling Li, and Heng Ji. 2024. Knowledge Overshadowing Causes Amalgamated Hallucination in Large Language Models. arXiv:2407.08039 [cs.CL] https://arxiv.org/abs/2407. 08039

    2024 arXiv

  25. [33]

    Yue Zhang, Yafu Li, Leyang Cui, Deng Cai, Lemao Liu, Tingchen Fu, Xinting Huang, Enbo Zhao, Yu Zhang, Yulong Chen, Longyue Wang, Anh Tuan Luu, Wei Bi, Freda Shi, and Shuming Shi. 2023. Siren’s Song in the AI Ocean: A Survey on Hallucination in Large Language Models. arXiv prep...

    2023 arXiv

  26. [34]

    Zhenliang Zhang, Xinyu Hu, Huixuan Zhang, Junzhe Zhang, and Xiaojun Wan

  27. [35]

    Jieyu Zhao, Tianlu Wang, Mark Yatskar, Vicente Ordonez, and Kai-Wei Chang

  28. [2018]

    arXiv preprint arXiv:1804.06876 (2018)

    Gender bias in coreference resolution: Evaluation and debiasing methods. arXiv preprint arXiv:1804.06876 (2018)

    2018 arXiv

  29. [2021]

    Transactions of the Association for Computational Linguistics 9 (2021), 845–874

    Gender Bias in Machine Translation. Transactions of the Association for Computational Linguistics 9 (2021), 845–874. doi:10.1162/tacl_a_00401

    2021 doi

  30. [2022]

    Survey of Hallucination in Natural Language Generation. Comput. Surveys 55 (2022), 1 – 38. https://api.semanticscholar.org/CorpusID:246652372

    2022

  31. [2024]

    arXiv:2406.11267 [cs.CL] https://arxiv.org/abs/2406.11267

    Mitigating Large Language Model Hallucination with Faithful Finetuning. arXiv:2406.11267 [cs.CL] https://arxiv.org/abs/2406.11267

    arXiv

  32. [2025]

    ICR Probe: Tracking Hidden State Dynamics for Reliable Hallucination Detection in LLMs. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , Wanxiang Che, Joyce Nabende, Ekaterina Shutova, and Mohammad Taher Pileh...

    2025 doi

  33. [3219]

    doi:10.18653/v1/2023.eacl-main.234

    2023 doi

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.