How Can We Know When Language Models Know? On the Calibration of Language Models for Question Answering
Open accessThe DOI matches its Crossref record; 72 of the 72 references it cites were found in a registry or an index.
Abstract
Recent works have shown that language models (LM) capture different types of knowledge regarding facts or common sense. However, because no model is perfect, they still fail to provide appropriate answers in many cases. In this paper, we ask the question, “How can we know when language models know, with confidence, the answer to a particular query?” We examine this question from the point of view of calibration, the property of a probabilistic model’s predicted probabilities actually being well correlated with the probabilities of correctness. We examine three strong generative models—T5, BART, and GPT-2—and study whether their probabilities on QA tasks are well calibrated, finding the answer is a relatively emphatic no. We then examine methods to calibrate such models to make their confidence scores correlate better with the likelihood of correctness through fine-tuning, post-hoc probability modification, or adjustment of the predicted outputs or inputs. Experiments on a diverse range of datasets demonstrate the effectiveness of our methods. We also perform analysis to study the strengths and limitations of these methods, shedding light on further improvements that may be made in methods for calibrating LMs. We have released the code at https://github.com/jzbjyb/lm-calibration.
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.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source doesn’t support the claim as written.
In the citing paper
“Another line of work that sheds light on the nature of hallucinations shows that LMs know when they’re hallucinating [1,3,18,20].”
From Calibrated Language Models Must Hallucinate (2024) · cites this paper as [18]
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measuredThe cited source doesn’t support the claim as written.
In the citing paper
“Jiang et al. [51] observed the probabilities of pre-trained LMs (e.g., BART, T5, and GPT-2) on QA tasks are well calibrated.”
From Pre-train, Prompt, and Predict: A Systematic Survey of Prompting Methods in Natural Language Processing (2023) · cites this paper as [51]
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Past work on uncertainty for LLMs has focused on simpler settings, such as classifiers16,17 and regressors18,19, whereas the most exciting applications of LLMs relate to free-form generations.”
From Detecting hallucinations in large language models using semantic entropy (2024) · cites this paper as [16]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“find only a weak correlation between the log-likelihood models assign to their answer and the answer's correctness.”
From Semantic Uncertainty: Linguistic Invariances for Uncertainty Estimation in Natural Language Generation (2023) · cites this paper as Jiang et al. (2021)
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“More complex decision-making is currently bottlenecked by issues such as over-confidence and miscalibration (Jiang et al., 2021; Braverman et al., 2020; Chen et al., 2022), misalignment with human values or bias (Liang et al., 2021; Feng et al., 2023), hallucinations in self-evaluation (Shinn et al., 2023), and lack of human-in-the-loop mechanisms in face of uncertainties (Nguyen et al., 2022a; R”
From Cognitive Architectures for Language Agents (2023) · cites this paper as Jiang et al., 2021
8 more claims about this paper
Create a free account to see every claim other papers make about this one.
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
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474 · cited 2×
“We found that larger LMs usually achieve both higher accuracy and better calibration performance, which is contradictory to the observation in image clas-sification (Guo et al., 2017) where larger models such as ResNet (He et al., 2016) are no longer well calibrated compared to smaller models like LeNet (Lecun et al.,…”
MethodVaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017) · arXiv · doi:10.48550/arxiv.1706.03762 · Cited by 120,404 · cited 1×
“The T5 model is a sequence-to-sequence model with both encoder and decoder using trans-formers (Vaswani et al., 2017), and the largest version has 11 billion parameters, allowing it to realize state-of-the-art performance on tasks such as question answering and natural language understanding…”
BackgroundGradient-based learning applied to document recognition
Lecun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998) · Proceedings of the IEEE · doi:10.1109/5.726791 · Cited by 58,805 · cited 2×
“We found that larger LMs usually achieve both higher accuracy and better calibration performance, which is contradictory to the observation in image clas-sification (Guo et al., 2017) where larger models such as ResNet (He et al., 2016) are no longer well calibrated compared to smaller models like LeNet (Lecun et al.,…”
MethodXGBoost: A Scalable Tree Boosting System
Chen, T., & Guestrin, C. (2016) · Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining · doi:10.1145/2939672.2939785 · Cited by 54,569 · cited 2×
“We follow previous works (Jagannatha and Yu, 2020; Dong et al., 2018) and use gradient boosted decision trees (Chen and Guestrin, 2016) as our regres-sor to estimate the confidence based on features.”
MethodInfluentialBERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019) · Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) · doi:10.18653/v1/n19-1423 · Cited by 54,325 · cited 3×
“LMs are now a ubiquitous tool in not only natu-ral language generation, but also natural language understanding (NLU), where they are largely used for unsupervised representation learning in pre-trained models such as BERT (Devlin et al., 2019).”
MethodInfluentialLanguage Models are Few-Shot Learners
Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al. (2020) · arXiv · doi:10.48550/arxiv.2005.14165 · Cited by 23,372
Kaugeranna, R., Kaugeranna, E., & Aion, (. S. 4. 6. (2023) · DROPS (Schloss Dagstuhl – Leibniz Center for Informatics) · doi:10.4230/lipics.giscience.2023.43 · Cited by 13,690
Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
Raffel, C., Shazeer, N., Roberts, A., Lee, K., Narang, S., Matena, M., Zhou, Y., Li, W., & Liu, P. J. (2019) · arXiv · doi:10.48550/arxiv.1910.10683 · Cited by 10,929 · cited 3×
“No-tably, large models trained on massive crawls of internet text (such as T5 (Raffel et al., 2019) and GPT-3 (Brown et al., 2020)) have been shown to be able to perform quite sophisticated knowledge-based tasks simply through prompting the model to predict the next words given a particular cue.”
BackgroundMethodScaling Laws for Neural Language Models
Kaplan, J., McCandlish, S., Henighan, T., Brown, T. B., Chess, B., Child, R., Gray, S., Radford, A., Wu, J., & Amodei, D. (2020) · arXiv · doi:10.48550/arxiv.2001.08361 · Cited by 9,550 · cited 1×
“Unlike ResNet trained on CIFAR-100, the training of LMs is not bottlenecked by the dataset, and larger LMs have a stronger capacity to model text distribution and memorize facts, which leads to better calibration performance overall (Kaplan et al., 2020).”
MethodLewis, M., Liu, Y., Goyal, N., Ghazvininejad, M., Mohamed, A., Levy, O., Stoyanov, V., & Zettlemoyer, L. (2020) · Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/2020.acl-main.703 · Cited by 8,821
Research works printed but not linked: 4
- Hewitt, 2019, A structural probe for finding syntax in word representations, 4129
- Ng, 2019, Facebook fair’s WMT19 news translation task submission, 314
- Quach, 2020, Researchers made an OpenAI GPT-3 medical chatbot as an experiment. It told a mock patient to kill themselves, The Register
- Radford, 2019, Language models are unsupervised multitask learners, OpenAI Blog, 1
Cited by
Items 101–200 of 307
Generative AI-Based Text Generation Methods Using Pre-Trained GPT 2 ModelPreprint
Pandey, R., Waghela, H., Rakshit, S., Rangari, A., Singh, A., Kumar, R., Ghoshal, R., & Sen, J. (2024) · doi:10.36227/techrxiv.171216659.95569463/v1 · Cited by 15
Epistemic Integrity in Large Language ModelsPreprint
Ghafouri, B., Mohammadzadeh, S., Zhou, J., Nair, P., Tian, J.-J., Tsujimura, H., Goel, M., Krishna, S., Rabbany, R., Godbout, J., et al. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2411.06528 · Cited by 22
“In tasks such as question-answering, internal certainty is often represented by the probability the model assigns to its selected response compared to alternative answers (Jiang et al., 2021; Hendrycks et al., 2021).”
Influentialvon Recum, A., Schnabl, C., Hollbeck, G., Alberti, S., Blinde, P., & Hagen, M. V. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2412.16974 · Cited by 20
“Jiang et al. [2021] show that models tend to perform poorly on questions that are unanswerable, also finding that prediction confidence in early transformer-based language models was not a reliable indicator of the model’s uncertainty.”
Rafe, A., & Das, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.24052 · Cited by 15
“Pre-trained transformers were found to be reasonably calibrated in domain out of the box (Desai and Durrett, 2020), and generative language models answering factual questions were found to be poorly calibrated (Jiang et al., 2021).”
Rafe, A., & Das, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.00346 · Cited by 22
“…are well calibrated in domain and less so out of domain relative to their in-domain level (Desai and Durrett, 2020), and the calibration of prompted language models, including whether their stated confidence tracks correctness, has its own track (Jiang et al., 2021; Kadavath et al., 2022).”
Knowledge of Knowledge: Exploring Known-Unknowns Uncertainty with Large Language Models
Amayuelas, A., Wong, K., Pan, L., Chen, W., & Wang, W. (2023) · arXiv · doi:10.48550/arxiv.2305.13712 · Cited by 15
Uncertainty Estimation in Large Language Models to Support Biodiversity Conservation
Mora-Cross, M. A., & Calderon-Ramirez, S. (2024) · Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT) · doi:10.18653/v1/2024.naacl-industry.31 · Cited by 22
“Incorporating uncertainty measures with predictions helps mitigate risks in decision-making and other applications (Kim et al., 2022; Jiang et al., 2021).”
BackgroundRethinking Error: “Hallucinations” and Epistemological Indifference
Fredrikzon, J. (2025) · Critical AI · doi:10.1215/2834703x-11700255 · Cited by 15
CHARD: Clinical Health-Aware Reasoning Across Dimensions for Text Generation Models
Feng, S. Y., Khetan, V., Sacaleanu, B., Gershman, A. V., & Hovy, E. H. (2023) · Conference of the European Chapter of the Association for Computational Linguistics · doi:10.18653/v1/2023.eacl-main.24 · Cited by 17
Wu, Q., Khan, M. A., Das, S., Nanda, V., Ghosh, B., Kolling, C., Speicher, T., Bindschaedler, L., Gummadi, K. P., & Terzi, E. (2025) · ACM International Conference on Web Search and Data Mining (WSDM) · doi:10.1145/3701551.3703562 · Cited by 22
“Reliance on LLMs’ meta-linguistic judgments: Prior works used prompt templates with instructions [7, 9, 19–21, 29, 33, 37, 39, 44, 49] for communicating the question as well as the expected format of answers.”
InfluentialAdaptation with Self-Evaluation to Improve Selective Prediction in LLMs
Chen, J., Yoon, J., Ebrahimi, S., Arik, S., Pfister, T., & Jha, S. (2023) · Findings of the Association for Computational Linguistics: EMNLP 2023 · doi:10.18653/v1/2023.findings-emnlp.345 · Cited by 13
Shen, K., & Kejriwal, M. (2025) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2025.112253 · Cited by 16
From text to multimodal: a survey of adversarial example generation in question answering systems
Yiğit, G., & Amasyalı, M. F. (2024) · Knowledge and Information Systems · doi:10.1007/s10115-024-02199-z · Cited by 13
Rethinking and recomputing the value of machine learning models
Sayin, B., Yang, J., Chen, X., Passerini, A., & Casati, F. (2025) · Artificial Intelligence Review · doi:10.1007/s10462-025-11242-6 · Cited by 15
Groot, T., & Valdenegro - Toro, M. (2024) · Proceedings of the 4th Workshop on Trustworthy Natural Language Processing (TrustNLP 2024) · doi:10.18653/v1/2024.trustnlp-1.13 · Cited by 20
Reliable Text-to-SQL with Adaptive Abstention
Chen, K., Chen, Y., Koudas, N., & Yu, X. (2025) · Proceedings of the ACM on Management of Data · doi:10.1145/3709719 · Cited by 12
Multi-Bin Batching for Increasing LLM Inference Throughput
Guldogan, O., Kunde, J., Lee, K., & Pedarsani, R. (2026) · ACM Transactions on Modeling and Performance Evaluation of Computing Systems · doi:10.1145/3843225 · Cited by 15
“Large Language Model (LLM) inference systems are becoming increasingly popular due to their various abilities, such as text generation (Li et al., 2024), coding assistance (Chen et al., 2021), and question answering (Jiang et al., 2021).”
On the attribution of confidence to large language models
Keeling, G., & Street, W. (2025) · Inquiry · doi:10.1080/0020174x.2025.2450598 · Cited by 13
“These results are troubling given that techniques for anticipating hallucinations and minimising the generation of misinformation by LLMs often rely upon eliciting calibrated confidence scores (Jiang et al. 2021; Varshney et al. 2023).”
Language Models Can Predict Their Own Behavior
Ashok, D., & May, J. (2025) · neural information processing systems · doi:10.52202/085713-2102 · Cited by 21
Cut Costs, Not Accuracy: LLM-Powered Data Processing with Guarantees
Zeighami, S., Shankar, S., & Parameswaran, A. (2025) · Proceedings of the ACM on Management of Data · doi:10.1145/3769776 · Cited by 30
Semantic-Oriented Unlabeled Priming for Large-Scale Language Models
Liu, Y., Schick, T., & Schtze, H. (2023) · Workshop on Simple and Efficient Natural Language Processing (SustaiNLP) · doi:10.18653/v1/2023.sustainlp-1.2 · Cited by 19
Selective-LAMA: Selective Prediction for Confidence-Aware Evaluation of Language Models
Yoshikawa, H., & Okazaki, N. (2023) · Findings of the Association for Computational Linguistics: EACL · doi:10.18653/v1/2023.findings-eacl.150 · Cited by 38
“Jiang et al. (2021) addressed a similar problem, but focused on a calibration of the model prediction on QA tasks.”
BackgroundLearning to Imagine: Visually-Augmented Natural Language Generation
Tang, T., Chen, Y., Du, Y., Li, J., Zhao, W. X., & Wen, J.-R. (2023) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2023.acl-long.526 · Cited by 12
Mo, G., Xing, Z., Zhu, Y., Huang, L., Chen, W., & Li, P. (2025) · Water · doi:10.3390/w17030354 · Cited by 11
Uncertainty Quantification for Text Classification
Zhang, D., Şensoy, M., Makrehchi, M., Taneva-Popova, B., Gui, L., & He, Y. (2023) · International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR) · doi:10.1145/3539618.3594243 · Cited by 15
“Furthermore, we will talk about the latest advances in uncertainty quantification for pre-trained language models (including asking language models to express their uncertainty [17, 24], interpreting uncertainties of text classifiers built on large-scale language models, uncertainty estimation in text generation [19],…”
BackgroundGenetic Prompt Search via Exploiting Language Model Probabilities
Zhao, J., Wang, Z., & Yang, F. (2023) · International Joint Conference on Artificial Intelligence · doi:10.24963/ijcai.2023/588 · Cited by 21
“Other proposed methodologies include mining prompt templates from the web [Jiang et al. , 2021] and training specific prompt generators [Ben-David et al. , 2022], which correspond to extra computational and human efforts.”
MethodSurvey of Large Language Models for Answering Questions Across Various Fields
Shailendra, P., Ghosh, R. C., Kumar, R., & Sharma, N. (2024) · International Conference on Advanced Computing and Communication Systems (ICACCS) · doi:10.1109/icaccs60874.2024.10717078 · Cited by 24
“Recent research suggests that large language models can grasp phrase syntax, factual information, and common knowledge and even perform basic mathematical operations [2].”
BackgroundOpen Relation Modeling: Learning to Define Relations between Entities
Huang, J., Chang, K. C., Xiong, J., & Hwu, W. (2022) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2022.findings-acl.26 · Cited by 14
Completeness, Recall, and Negation in Open-world Knowledge Bases: A Survey
Razniewski, S., Arnaout, H., Ghosh, S., & Suchanek, F. (2024) · ACM Computing Surveys · doi:10.1145/3639563 · Cited by 8
“Knowledge extraction from pre-trained language models has recently received much attention [87], yet it remains unclear to which degree this approach can yield knowledge for multi-valued, optional, and long-tail predicates [56, 112].”
BackgroundConformal Autoregressive Generation: Beam Search with Coverage Guarantees
Deutschmann, N., Alberts, M., & Martínez, M. R. (2024) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v38i10.29062 · Cited by 25
Pre‐trained language models: What do they know?
Guimar�ães, N., Campos, R., & Jorge, A. (2023) · Wiley Interdisciplinary Reviews Data Mining and Knowledge Discovery · doi:10.1002/widm.1518 · Cited by 17
“In fact, prompt engineering is a rapidly growing field, with different prompt architectures being studied and developed and with empirical results showing that performance largely depends on the prompts provided (Jiang et al., 2021).”
BackgroundLLM Driven Web Profile Extraction for Identical Names
Sancheti, P., Karlapalem, K., & Vemuri, K. (2024) · International World Wide Web Conference (WWW) · doi:10.1145/3589335.3651946 · Cited by 15
“LLMs tend to confidently identify empty or null inputs as important entities [8, 15, 28].”
Steering Large Language Models for Cross-lingual Information Retrieval
Guo, P., Ren, Y., Hu, Y., Cao, Y., Li, Y., & Huang, H. (2024) · International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR) · doi:10.1145/3626772.3657819 · Cited by 25
ELQA: A Corpus of Metalinguistic Questions and Answers about English
Behzad, S., Sakaguchi, K., Schneider, N., & Zeldes, A. (2023) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2023.acl-long.113 · Cited by 19
“We believe that answer confidence estimation—so that users can be shown only the best model-generated answers—is an important direction for using learner QA models in practice (Jiang et al., 2021).”
BackgroundEvaluating the Knowledge Base Completion Potential of GPT
Veseli, B., Razniewski, S., Kalo, J.-C., & Weikum, G. (2023) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2023.findings-emnlp.426 · Cited by 9
“Further research on calibrating model confidences seems important (Jiang et al., 2021; Singhania et al., 2023).”
BackgroundAmbiCoref: Evaluating Human and Model Sensitivity to Ambiguous Coreference
Yuan, Y., Malaviya, C., & Yatskar, M. (2023) · Findings of the Association for Computational Linguistics: EACL · doi:10.18653/v1/2023.findings-eacl.75 · Cited by 9
Watson, W., Cho, N., & Srishankar, N. (2025) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v39i24.34736 · Cited by 13
“Despite the promising potential for a myriad of use cases, Large Language Models (LLMs) offer limited insights into their chain of thought (Liang et al. 2022; Wei et al. 2023; Kojima et al. 2023) and have the propensity to hallucinate in various circumstances (Jiang et al. 2021).”
An Analysis of Social Biases Present in BERT Variants Across Multiple LanguagesPreprint
Milios, A., & BehnamGhader, P. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2211.14402 · Cited by 11
Do Large Language Models have Shared Weaknesses in Medical Question Answering?Preprint
Bean, A. M., Korgul, K., Krones, F., McCraith, R., & Mahdi, A. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2310.07225 · Cited by 12
I Don't Know: Explicit Modeling of Uncertainty with an [IDK] Token
Cohen, R., Dobler, K., Biran, E., & de Melo, G. (2024) · neural information processing systems · doi:10.52202/079017-0349 · Cited by 41
Task Cascades for Efficient Unstructured Data Processing
Shankar, S., Zeighami, S., & Parameswaran, A. (2026) · Proceedings of the ACM on Management of Data · doi:10.1145/3786702 · Cited by 13
“However, LLM-generated confidence scores are often poorly calibrated and heavily concentrated near 1, making uniform decrements ineffective [14, 29, 32].”
Extraction of Subjective Information from Large Language Models
Kobayashi, A., & Yamaguchi, S. (2024) · IEEE Annual International Computer Software and Applications Conference (COMPSAC) · doi:10.1109/compsac61105.2024.00253 · Cited by 8
“…that LLM benchmarks are lowered when manually generated hole-filling prompts are suboptimal, Zhengbao et al. proposed mining-based and paraphrase-based methods to automatically generate high-quality and diverse prompts, as well as ensemble methods to combine answers from different prompts [11].”
BackgroundAquaPipe: A Quality-Aware Pipeline for Knowledge Retrieval and Large Language Models
Yu, R., Huang, W., Bai, S., Zhou, J. B., & Wu, F. (2025) · Proceedings of the ACM on Management of Data · doi:10.1145/3709661 · Cited by 18
“First, Adaptive Retrieval determines the necessity of retrieval based on rules, ensuring that external information is retrieved only when essential, thereby reducing unnecessary retrieval [8, 16, 27, 32, 37, 46, 61].”
Fan, C., Tian, J., Li, Y., Chen, W., He, H., & Jin, Y. (2023) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2023.emnlp-main.913 · Cited by 11
Quantized Prompt for Efficient Generalization of Vision-Language Models
Hao, T., Ding, X., Feng, J., Yang, Y., Chen, H., & Ding, G. (2024) · Lecture notes in computer science · doi:10.1007/978-3-031-72655-2_4 · Cited by 19
Beyond Calibration: Do a Typed-Decision Model's Probabilities Obey the Probability Axioms?Preprint
Li, K., He, Y., & Li, Q. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.33209 · Cited by 10
Refining medical large language models: key insights from instruction tuning
Alqahtani, M. Q., Al-Barakati, A. A., & Alotaibi, F. S. (2025) · PeerJ Computer Science · doi:10.7717/peerj-cs.3216 · Cited by 7
Large Language Models: A Paradigm Shift for Dementia Diagnosis and Care
Harrison, J. R., Tang, S. L., Burston, B., Robertson, A., Liang, H., & Taylor, J. (2025) · British Journal of Hospital Medicine · doi:10.12968/hmed.2024.0666 · Cited by 9
From local to global: Leveraging document graph for named entity recognition
Shang, Y.-M., Mao, H., Tian, T., Huang, H., & Mao, X.-L. (2025) · Knowledge-Based Systems · doi:10.1016/j.knosys.2025.113017 · Cited by 13
Life after BERT: What do Other Muppets Understand about Language?
Lialin, V., Zhao, K., Shivagunde, N., & Rumshisky, A. (2022) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2022.acl-long.227 · Cited by 10
“Fortunately, this trend is changing and now we see more papers that probe models such as ALBERT , T5 or BART (Mosbach et al., 2020; Phang et al., 2021; Jiang et al., 2021).”
MethodThe Battleship Approach to the Low Resource Entity Matching Problem
Genossar, B., Gal, A., & Shraga, R. (2023) · Proceedings of the ACM on Management of Data · doi:10.1145/3626711 · Cited by 11
“However, highly parameterized models such as transformer-based pre-trained language models tend to produce an uncalibrated confidence value [17, 22], assigning mostly dichotomous values close to either 0 or 1.”
BackgroundMethodRoohi, S., Skarbez, R., & Nguyen, H. D. (2024) · Natural language processing. · doi:10.1017/nlp.2024.48 · Cited by 10
Deuce : Dual-diversity Enhancement and Uncertainty-awareness for Cold-start Active Learning
Guo, J., Chen, C. L. P., Li, S., & Zhang, T. (2024) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00731 · Cited by 13
Sugan, M., Nadkarni, S., Sanaeipoor, S., Patel, S., Natarajamoorthy, S. Y., Rashidi, E., & Miller, A. (2025) · CHI Conference Extended Abstracts on Human Factors in Computing Systems (CHI EA) · doi:10.1145/3706599.3721270 · Cited by 6
Learning Shortcuts: On the Misleading Promise of NLU in Language Models
Bihani, G., & Rayz, J. T. (2025) · Springer eBooks · doi:10.1007/978-3-031-67886-8_9 · Cited by 8
Generative AI as a Supportive Tool for Scientific Research
Weinberg, A. I. (2024) · Springer eBooks · doi:10.1007/978-3-031-46238-2_1 · Cited by 8
A large-scale evaluation of commonsense knowledge in humans and large language models
Nguyen, T. D., Watts, D. J., & Whiting, M. E. (2026) · PNAS Nexus · doi:10.1093/pnasnexus/pgag029 · Cited by 10
Estimating Large Language Model Capabilities without Labeled Test Data
Fu, H. Y., Ye, Q., Xu, A., Ren, X., & Jia, R. (2023) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2023.findings-emnlp.639 · Cited by 6
Uncertainty Quantification in Retrieval Augmented Question Answering
Perez-Beltrachini, L., & Lapata, M. (2025) · Transactions on machine learning research · doi:10.48550/arxiv.2502.18108 · Cited by 10
“…have proven beneficial in increasing LLM performance on tail knowledge (Izacard et al., 2024; Mallen et al., 2023), reducing hallucinations in the generated answers, and even improving model cali-1 Code and data are available at https://github.com/lauhaide/ragu bration (Jiang et al., 2021).”
Inv-Entropy: A Fully Probabilistic Framework for Uncertainty Quantification in Language Models
Song, H., Ji, R., Shi, N., Lai, F., & Kontar, R. A. (2025) · neural information processing systems · doi:10.52202/085713-3598 · Cited by 5
SPE: Symmetrical Prompt Enhancement for Fact Probing
Li, Y., Che, T., Wang, Y., Jiang, Z., Xiong, C., & Chaturvedi, S. (2022) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2022.emnlp-main.803 · Cited by 8
Self-citationCalibration Error Estimation Using Fuzzy Binning
Bihani, G., & Rayz, J. T. (2023) · Lecture notes in networks and systems · doi:10.1007/978-3-031-46778-3_9 · Cited by 5
How Often Are Errors in Natural Language Reasoning Due to Paraphrastic Variability?
Srikanth, N. P., Carpuat, M. J., & Rudinger, R. (2024) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00692 · Cited by 10
Meng, X., Song, Y., Bai, Q., & Wang, T. (2023) · Knowledge-Based Systems · doi:10.1016/j.knosys.2023.110796 · Cited by 6
Generative AI-Based Text Generation Methods Using Pre-Trained GPT-2 ModelPreprint
Pandey, R. K., Waghela, H., Rakshit, S., Rangari, A., Singh, A., Kumar, R., Ghoshal, R., & Sen, J. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2404.01786 · Cited by 7
CyberNER-LLM: Cyber Threat Intelligence Named Entity Recognition With Large Language Model
Liu, X., Lin, W., & Ding, Z. (2026) · Lecture Notes in Computer Science · doi:10.1007/978-981-95-3543-9_28 · Cited by 4
Steer Your Model: Secure Code Generation With Contrastive Decoding
Huang, L., Yan, M. L., Yin, T., Sun, W., Liu, Z., Zhang, H., & Lo, D. (2026) · IEEE Transactions on Software Engineering · doi:10.1109/tse.2025.3650127 · Cited by 9
“Following Jiang et al. [38], we first estimate the model’s confidence from functional tuning as Similarly, the confidence after the positive security steering is estimated as Then, α is determined according to: The rationale is that a high confidence in the function-tuned predictions U ft signals the…”
Human Evaluation of Large Language Models: A Review and Protocol Selection Framework
Brunyé, T. T. (2026) · AI · doi:10.3390/ai7050174 · Cited by 4
Li, B., Liu, T., Ma, J., Dai, J., & Liao, P. (2025) · Computer Networks · doi:10.1016/j.comnet.2025.111074 · Cited by 5
Multimodal VAD: Visual Anomaly Detection in Intelligent Monitoring System via Audio-Vision-Language
Wang, D., Wang, Q., Hu, Q., & Wu, K. (2025) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2025.3578702 · Cited by 4
“In recent years, vision-language pretraining models have not only demonstrated remarkable progress in tasks, such as image captioning [37], visual question answering [38], and image-text matching [39], [40], but have also played a crucial role in industrial monitoring and measurement.”
Guiding Reinforcement Learning Using Uncertainty-Aware Large Language Models
Shoaeinaeini, M., & Harrison, B. E. (2025) · IEEE International Conference on Trust, Privacy and Security in Intelligent Systems and Applications (TPS-ISA) · doi:10.1109/tps-isa67132.2025.00045 · Cited by 4
“Deterministic methods like logit-based methods (log-probabilities) [49], [50] or entropy-based methods [23] are mainly applied in classification tasks.”
When to Read Documents or QA History: On Unified and Selective Open-domain QA
Lee, K., Han, S.-E., Hwang, S.-W., & Lee, M. (2023) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2023.findings-acl.401 · Cited by 5
What Has Been Enhanced in my Knowledge-Enhanced Language Model?
Hou, Y., Fu, G., & Sachan, M. (2022) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2022.findings-emnlp.102 · Cited by 6
“However, people have to manually design templates for factual knowledge to be probed5, and the quality of templates is vital in the overall prompt accuracy (Jiang et al., 2021; Li et al., 2022).”
BackgroundGuo, P., Hu, Y., Cao, Y., Ren, Y., Li, Y., & Huang, H. (2024) · International World Wide Web Conference (WWW) · doi:10.1145/3589334.3645701 · Cited by 6
A Study on the Calibrated Confidence of Text Classification Using a Variational Bayes
Lee, J., & Park, S. (2022) · Applied Sciences · doi:10.3390/app12189007 · Cited by 4
Trust but Verify: Prover-Verifier Deliberation for Selective LLM PredictionPreprint
Sedoc, J., Zhang, B., & Foster, D. P. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.25133 · Cited by 5
“Neural networks are often miscalibrated [10], and generative language models introduce additional calibration difficulties because probability mass is distributed over many possible completions [12].”
Song, P. (2025) · Procedia Computer Science · doi:10.1016/j.procs.2025.04.256 · Cited by 3
Schönwälder, E., Falkenberg, C., Hartmann, C., & Lehner, W. (2025) · IEEE International Conference on Data Science and Advanced Analytics (DSAA) · doi:10.1109/dsaa65442.2025.11248029 · Cited by 5
A statistically consistent measure of semantic uncertainty using Language Models
Liu, Y. (2026) · doi:10.1109/icnlp69856.2026.11527557 · Cited by 3
“Most prior studies on uncertainty in foundation models for natural language processing (NLP) have focused primarily on the calibration of classifiers and text regressors (Jiang et al., 2021; Desai and Durrett, 2020; Glushkova et al., 2021).”
Zou, J., Qin, Y., & Rhee, M. (2024) · Innovation and technology studies. · doi:10.61784/its4026 · Cited by 27
Graph-Enhanced Prompt Learning for Cross-Domain Contract Element Extraction
Wang, Z., Wang, H., Ren, P., Chen, Z., de Rijke, M., & Ren, Z. (2025) · ACM Transactions on Information Systems · doi:10.1145/3715100 · Cited by 4
“Besides, prompts also enable knowledge probing to quantify knowledge presented in the LLMs for the specific tasks of interest [23, 29, 38, 54, 55, 80].”
Calibration Meets Explanation: A Simple and Effective Approach for Model Confidence Estimates
Li, D., Hu, B., & Chen, Q. (2022) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2022.emnlp-main.178 · Cited by 11
Made of Steel? Learning Plausible Materials for Components in the Vehicle Repair Domain
Eichel, A., Schlipf, H., & Walde, S. S. I. (2023) · Conference of the European Chapter of the Association for Computational Linguistics · doi:10.18653/v1/2023.eacl-main.104 · Cited by 7
Bako, H. K., Bhutani, A., Liu, X., Cobbina, K. A., & Liu, Z. (2024) · Visualization · doi:10.1109/vis55277.2024.00060 · Cited by 3
Conformal Language Model Reasoning with Coherent Factuality
Rubin-Toles, M., Gambhir, M., Ramji, K., Roth, A., & Goel, S. (2025) · International Conference on Learning Representations · doi:10.48550/arxiv.2505.17126 · Cited by 6
“Prior works have explored expressions of uncertainty including logit weights (Guo et al., 2017; Jiang et al., 2021), surrogate estimates (Shrivastava et al., 2023), sampling variance (Kuhn et al., 2023; Xiong et al., 2024), and natural language generations indicating uncertainty (Lin et al., 2022;…”
Multicalibration for LLM-based Code GenerationPreprint
Campos, V., Kuschnereit, R., & Ulges, A. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2512.08810 · Cited by 3
“A key challenge in our setup is selecting an appropriate scoring method to derive initial confidences ˆ 𝑝 as a starting point for calibration.”
Confidence Regulation Neurons in Language Models
Stolfo, A., Wu, B., Gurnee, W., Belinkov, Y., Song, X., Sachan, M., & Nanda, N. (2024) · Advances in Neural Information Processing Systems 37 · doi:10.52202/079017-3970 · Cited by 2
The Benefits of Query-Based KGQA Systems for Complex and Temporal Questions in LLM Era
Алексеев, А. А., Chaichuk, M., Butko, M., Panchenko, A., Tutubalina, E. V., & Somov, O. (2025) · Lecture notes in computer science · doi:10.1007/978-3-031-97141-9_29 · Cited by 2
“Thus LLM-based QA systems often struggle with answering rare, multi-hop, or time-sensitive questions – especially those requiring structured data, despite the possibility of answering such question with internal or external storage [18].”
Abumelha, M., ALGhamdi, A. S. A.-M., Fayoumi, A. G., & Ragab, M. (2025) · JMIR Medical Informatics · doi:10.2196/78432 · Cited by 4
“We achieve this through a customized loss function that penalizes overconfident incorrect predictions, drawing on previous work in uncertainty estimation [22,45].”
Carou‐Senra, P., Rodríguez‐Pombo, L., Alvarez‐Lorenzo, C., & Goyanes, Á. (2026) · International Journal of Pharmaceutics · doi:10.1016/j.ijpharm.2026.126587 · Cited by 3
LADA: Label Disambiguation and Domain-Aware Learning for Domain Generalization
Xiao, Z. J., Wang, H., Ye, Y., Ye, W., Chen, H., Chen, G., Zhao, J., & Ying, R. (2026) · Machine Learning · doi:10.1007/s10994-025-06977-w · Cited by 2
“An explicit observation is that the choice of input prompts is crucial for the zero-shot performance (Zhong et al., 2021; Jiang et al., 2021; Shin et al., 2020).”
Uncertainty Calibration in Deep Learning: Methods, Emerging Challenges, and LLM Frontiers
Zhang, M.-L., & Wang, D.-B. (2026) · Journal of Computer Science and Technology · doi:10.1007/s11390-026-6426-z · Cited by 8
Evaluating LLM’s Code Reading Abilities in Big Data Contexts using Metamorphic Testing
Li, Z., Li, Z., Xiao, K., & Li, X. (2023) · International Conference on Big Data and Information Analytics (BigDIA) · doi:10.1109/bigdia60676.2023.10429345 · Cited by 2
“…formatted data and unformatted data and are capable of performing a wide variety of natural language processing tasks, including text generation [6, 7], translation [1, 8], summarization [9, 10], question answering [11, 12], and code generation and comprehension [13, 14]. text generation [6, 7].”
BackgroundNeurosymbolic Information Extraction from Transactional Documents
Hemmer, A., Coustaty, M., Bartolo, N., & Ogier, J.-M. (2025) · International Journal on Document Analysis and Recognition (IJDAR) · doi:10.1007/s10032-025-00530-0 · Cited by 2
“However, this progress has come with its own set of problems: LLMs lack symbolic reasoning capabilities [11], can confidently generate incorrect information [12, 13], and require substantial computational resources and data for training and inference [14].”
Do LLMs Understand Collaborative Signals? Diagnosis and Repair
Pouryousef, S., & Montazeralghaem, A. (2026) · Conference on Human Information Interaction and Retrieval · doi:10.1145/3786304.3787876 · Cited by 3
“Large Language Models (LLMs) [1, 2] have demonstrated impressive capabilities in a wide range of tasks, including reasoning over textual input [3, 4], answering complex questions [5, 6], generating fluent text [7], and encoding world knowledge [8].”
Trustworthy machine reading comprehension with conditional adversarial calibration
Wu, Z., & Xu, H. (2022) · Applied Intelligence · doi:10.1007/s10489-022-04235-3 · Cited by 2
“On the one hand, some work estimates the uncertainty in unsupervised way, like softmax-based and ensembled based methods [16, 17, 19, 20, 31].”
BackgroundMethodRobust Representation Learning
Cui, G., Liu, Z., Lin, Y., & Sun, M. (2023) · Springer eBooks · doi:10.1007/978-981-99-1600-9_8 · Cited by 2
Multimodal Shannon Game with Images
Zouhar, V., Bhattacharya, S., & Bojar, O. (2025) · Lecture notes in computer science · doi:10.1007/978-3-031-82487-6_3 · Cited by 2
Context-Aware Differential Privacy for Language ModelingPreprint
Dinh, M. H., & Fioretto, F. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2301.12288 · Cited by 5
“, xi−1) into φ(x) that satisfies the α-context notion introduced above, using a round trip translation strategy [11].”
MethodQuantifying Uncert-AI-nty: Testing the Accuracy of LLMs’ Confidence JudgmentsPreprint
Cash, T. N., Oppenheimer, D. M., Christie, S., & Devgan, M. (2025) · PsyArXiv (OSF Preprints) · doi:10.31234/osf.io/47df5_v4 · Cited by 2
Metadata
- Authors
- Jiang, Z., Araki, J., Ding, H., & Neubig, G.
- Affiliations
- Carnegie Mellon University (US)
- Robert Bosch (United States) (US)
- Corresponding author
- Jiang, Zhengbao
- Araki, Jun
- Ding, Haibo
- Neubig, Graham
- Venue
- Transactions of the Association for Computational Linguistics
- Publisher
- MIT Press
- Published
- 2021
- Volume
- 9
- Pages
- 962-977
- ISSN
- 2307-387X (ISSN-L)
- OpenAlex
- W3199958362
- Cited by
- 177
- References
- 71
- Citations by year
- 2026: 43 · 2025: 49 · 2024: 32 · 2023: 37 · 2022: 14 · 2021: 2
- FWCI
- 7.48
- Type
- journal article
- Language
- en
- Open access
- diamond
- doi.org
- Locations
- Transactions of the Association for Computational Linguistics (publishedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Artificial Intelligence › Adversarial Robustness in Machine Learning
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Keywords
- question answering, confidence calibration, probabilistic reasoning, T5, BART, GPT-2, fine-tuning, knowledge capture, factual knowledge, commonsense knowledge
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.