Understanding and Detecting Hallucinations in Neural Machine Translation via Model Introspection
Open accessThe DOI matches its Crossref record; 19 of the 19 references it cites were found in a registry or an index.
Abstract
Neural sequence generation models are known to “hallucinate”, by producing outputs that are unrelated to the source text. These hallucinations are potentially harmful, yet it remains unclear in what conditions they arise and how to mitigate their impact. In this work, we first identify internal model symptoms of hallucinations by analyzing the relative token contributions to the generation in contrastive hallucinated vs. non-hallucinated outputs generated via source perturbations. We then show that these symptoms are reliable indicators of natural hallucinations, by using them to design a lightweight hallucination detector which outperforms both model-free baselines and strong classifiers based on quality estimation or large pre-trained models on manually annotated English-Chinese and German-English translation test beds.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜
Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021) · Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency · doi:10.1145/3442188.3445922 · Cited by 6,988 · cited 1×
“, 2021), the output is completely detached from the source, which not only reveals fundamental limitations of current models, but also risks misleading users and undermining trust (Bender et al., 2021).”
BackgroundUnsupervised Cross-lingual Representation Learning at Scale
Conneau, A., Khandelwal, K., Goyal, N., Chaudhary, V., Wenzek, G., Guzmán, F., Grave, E., Ott, M., Zettlemoyer, L., & Stoyanov, V. (2020) · Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/2020.acl-main.747 · Cited by 5,637
On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer-Wise Relevance Propagation
Bach, S., Binder, A., Montavon, G., Klauschen, F., Müller, K.-R., & Samek, W. (2015) · PLOS ONE · doi:10.1371/journal.pone.0130140 · Cited by 4,524 · cited 12×
“We test the three source contribution hypotheses described in § 2 on the resulting dataset by contrasting the contributions of relevant tokens to the generation of a hallucinated versus a nonhallucinated translation using LRP (Bach et al., 2015).”
BackgroundMethodInfluentialThe Curious Case of Neural Text Degeneration
Holtzman, A., Buys, J., Du, L., Forbes, M., & Choi, Y. (2019) · arXiv · doi:10.48550/arxiv.1904.09751 · Cited by 1,304
On Faithfulness and Factuality in Abstractive Summarization
Maynez, J., Narayan, S., Bohnet, B., & McDonald, R. (2020) · Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/2020.acl-main.173 · Cited by 1,185
On Faithfulness and Factuality in Abstractive Summarization
Maynez, J., Narayan, S., Bohnet, B., & McDonald, R. (2020) · arXiv · doi:10.48550/arxiv.2005.00661 · Cited by 1,185 · cited 1×
“Hallucinations occur in all applications of neural models to language generation, including abstractive summarization (Falke et al., 2019; Maynez et al., 2020b), dialogue generation (Dušek et al.”
BackgroundChallenges in Data-to-Document Generation
Wiseman, S., Shieber, S., & Rush, A. (2017) · Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d17-1239 · Cited by 524
Durmus, E., He, H., & Diab, M. (2020) · Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/2020.acl-main.454 · Cited by 306
Sharp nearby, fuzzy far away: How neural language models use context
Khandelwal, U., He, H., Qi, P., & Jurafsky, D. (2018) · Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/p18-1027 · Cited by 303
Sharp Nearby, Fuzzy Far Away: How Neural Language Models Use Context
Khandelwal, U., He, H., Qi, P., & Jurafsky, D. (2018) · arXiv · doi:10.48550/arxiv.1805.04623 · Cited by 302
Cited by
Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models
Dahl, M., Magesh, V., Suzgun, M., & Ho, D. E. (2024) · Journal of Legal Analysis · doi:10.1093/jla/laae003 · Cited by 559
“This is a major concern in tasks requiring a high degree of accuracy between the response and a long-form input, such as machine translation (Xu et al., 2023) or summarization (Cao et al., 2018).”
Towards trustworthy LLMs: a review on debiasing and dehallucinating in large language models
Lin, Z., Guan, S., Zhang, W., Zhang, H., Li, Y., & Zhang, H. (2024) · Artificial Intelligence Review · doi:10.1007/s10462-024-10896-y · Cited by 141
Hallucinations in Large Multilingual Translation Models
Guerreiro, N. M., Alves, D. M., Waldendorf, J., Haddow, B., Birch, A., Colombo, P., & Martins, A. F. T. (2023) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00615 · Cited by 124
Hallucinations in Large Multilingual Translation Models
Guerreiro, N. M., Alves, D., Waldendorf, J., Haddow, B., Birch, A., Colombo, P., & Martins, A. F. T. (2023) · arXiv · doi:10.48550/arxiv.2303.16104 · Cited by 124
Wang, F., Zhang, Z., Zhang, X., Wu, Z., Mo, T., Lu, Q., Wang, W., Li, R., Xu, J., Tang, X., et al. (2025) · ACM Transactions on Intelligent Systems and Technology · doi:10.1145/3768165 · Cited by 106
Calibrated Language Models Must Hallucinate
Kalai, A. T., & Vempala, S. S. (2024) · Proceedings of the 56th Annual ACM Symposium on Theory of Computing · doi:10.1145/3618260.3649777 · Cited by 91
Benchmarking Uncertainty Quantification Methods for Large Language Models with LM-Polygraph
Vashurin, R., Fadeeva, E., Vazhentsev, A., Rvanova, L., Vasilev, D., Tsvigun, A., Petrakov, S., Xing, R., Sadallah, A., Grishchenkov, K., et al. (2025) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00737 · Cited by 21
“For instance, fact-checkers target a very narrow sub-problem, and usually require ex-ternal knowledge sources such as knowledge bases that are generally incomplete.”
LM-Polygraph: Uncertainty estimation for language models
Fadeeva, E., Vashurin, R., Tsvigun, A., Vazhentsev, A., Petrakov, S., Fedyanin, K., Vasilev, D., Goncharova, E., Panchenko, A., Panov, M., et al. (2023) · Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing: System Demonstrations · doi:10.18653/v1/2023.emnlp-demo.41 · Cited by 20
From bias to black boxes: understanding and managing the risks of AI – an actuarial perspective
du Preez, V., Bennet, S., Byrne, M., Couloumy, A., Das, A., Dessain, J., Galbraith, R., King, P., Mutanga, V., Schiller, F., et al. (2024) · British Actuarial Journal · doi:10.1017/s1357321724000060 · Cited by 15
Fine-tuning Language Models for Factuality
Tian, K., Mitchell, E., Yao, H., Manning, C. D., & Finn, C. (2023) · arXiv · doi:10.48550/arxiv.2311.08401 · Cited by 14
Multiagent Actor-Critic Generative AI for Query Resolution and Analysis
Ur Rahman, M. W., Nevarez, R., Mim, L. T., & Hariri, S. (2025) · IEEE Transactions on Artificial Intelligence · doi:10.1109/tai.2025.3544173 · Cited by 8
Ning, W., Li, M., Reimers, J. R., & Kobayashi, R. (2026) · Digital Discovery · doi:10.1039/d5dd00482a · Cited by 8
Metadata
- Authors
- Xu, W., Agrawal, S., Briakou, E., Martindale, M. J., & Carpuat, M.
- Venue
- Transactions of the Association for Computational Linguistics
- Publisher
- MIT Press
- Published
- 2023
- Volume
- 11
- Pages
- 546-564
- ISSN
- 2307-387X (ISSN-L)
- OpenAlex
- W4380995299
- Cited by
- 53
- References
- 19
- FWCI
- 4.66
- Type
- journal article
- Language
- en
- Open access
- diamond
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.