Benchmarking Large Language Models for News Summarization
Open accessThe DOI matches its Crossref record; 86 of the 86 references it cites were found in a registry or an index.
Abstract
Large language models (LLMs) have shown promise for automatic summarization but the reasons behind their successes are poorly understood. By conducting a human evaluation on ten LLMs across different pretraining methods, prompts, and model scales, we make two important observations. First, we find instruction tuning, not model size, is the key to the LLM’s zero-shot summarization capability. Second, existing studies have been limited by low-quality references, leading to underestimates of human performance and lower few-shot and finetuning performance. To better evaluate LLMs, we perform human evaluation over high-quality summaries we collect from freelance writers. Despite major stylistic differences such as the amount of paraphrasing, we find that LLM summaries are judged to be on par with human written summaries.
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 source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“For summarization tasks, although LLMs do not have an obvious advantage over fine-tuned models under traditional automatic evaluation metrics, such as ROUGE [70], human evaluation results indicate that humans tend to prefer the results generated by LLMs [42,138] compared to that of fine-tuned models.”
From Harnessing the power of llms in practice: a survey on chatGPT and beyond (2024) · cites this paper as [138]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“LLMs are known to produce errors in many settings, from document summarisation7, to general reasoning tasks as well as more clinically relevant tasks8.”
From A framework to assess clinical safety and hallucination rates of LLMs for medical text summarisation (2025) · cites this paper as [7]
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
Language 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 · cited 2×
“LLMs (Bommasani et al., 2021; Chowdhery et al., 2022; Brown et al., 2020) have two distinctive features over previous pretrained models.”
BackgroundSequence to Sequence Learning with Neural Networks
Sutskever, I., Vinyals, O., & Le, Q. V. (2014) · arXiv · doi:10.48550/arxiv.1409.3215 · Cited by 17,766 · cited 1×
“In this work, we evaluate the effectiveness of automatic metrics for evaluating LLMs and show that the usefulness of reference-based evaluation is closely linked to the quality of the references.”
MethodNeural Machine Translation by Jointly Learning to Align and Translate
Bahdanau, D., Cho, K., & Bengio, Y. (2014) · arXiv · doi:10.48550/arxiv.1409.0473 · Cited by 16,213 · cited 1×
“In this work, we evaluate the effectiveness of automatic metrics for evaluating LLMs and show that the usefulness of reference-based evaluation is closely linked to the quality of the references.”
Method(2004) · Cited by 9,718
Lewis, 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
Lewis, M., Liu, Y., Goyal, N., Ghazvininejad, M., Mohamed, A., Levy, O., Stoyanov, V., & Zettlemoyer, L. (2019) · arXiv · doi:10.48550/arxiv.1910.13461 · Cited by 8,816 · cited 2×
“In this work, we evaluate the effectiveness of automatic metrics for evaluating LLMs and show that the usefulness of reference-based evaluation is closely linked to the quality of the references.”
MethodTraining language models to follow instructions with human feedback
Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C. L., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., et al. (2022) · arXiv · doi:10.48550/arxiv.2203.02155 · Cited by 4,208 · cited 2×
“In this approach, a diverse set of natural language processing tasks are reformulated into the prompting format and the LLM’s parameters are updated for these tasks either through supervised finetuning or reinforcement learning.”
Method(2005) · Cited by 4,152
Get To The Point: Summarization with Pointer-Generator Networks
See, A., Liu, P. J., & Manning, C. D. (2017) · Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/p17-1099 · Cited by 3,775 · cited 2×
“In this work, we evaluate the effectiveness of automatic metrics for evaluating LLMs and show that the usefulness of reference-based evaluation is closely linked to the quality of the references.”
BackgroundMethodTraining Language Models to Follow Instructions with Human Feedback
Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., et al. (2022) · Advances in Neural Information Processing Systems 35 · doi:10.52202/068431-2011 · Cited by 3,117 · cited 2×
“In this approach, a diverse set of natural language processing tasks are reformulated into the prompting format and the LLM’s parameters are updated for these tasks either through supervised finetuning or reinforcement learning.”
Method
Research works printed but not linked: 1
- Conroy, 2006, Back to basics: Classy 2006, 10.12968/sece.2006.11.755s · doi:10.12968/sece.2006.11.755s
Cited by
Items 401–427 of 427
AutoChecklist: Automated Checklist Refinement for LLM Judges
Uniyal, M., Singh, M. P., Verbruggen, G., Le, V., & Gulwani, S. (2026) · ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE) · doi:10.1145/3803437.3805263
“Fromcustomer-facingchatbotstointernalautomation,organizationsrelyonLLMs fortasksliketextsummarization[24],machinetranslation[23], questionanswering[9],andcontentgeneration[16,21].”
RAGnRoll: Learning to Iteratively Retrieve and Generate Attributable Answer Snippets
Djeddal, H., Soulier, L., Pinel-Sauvagnat, K., Katrenko, S., Erbacher, P., & Tamine, L. (2026) · ACM Transactions on Information Systems · doi:10.1145/3831679
“Recent advances in Large Language Models (LLMs) [12, 36] have significantly enhanced their capabilities in language understanding and generation [23, 104, 107], establishing them as powerful tools for a wide range of tasks that extend beyond the scope of traditional Natural Language Processing…”
Making Lectures Accessible: AI-Based Abstractive Summarisation for Video Content
Martinez, H. B., Walsh, M., Fagan, C., & Hines, A. (2026) · Technology Knowledge and Learning · doi:10.1007/s10758-026-10015-6
ConfigTransLE: Large Language Models Enhanced Network Configuration Translation
Li, F., Zheng, N., Liang, B., Yang, Y., Liu, C., Gao, C., Wang, X., & Cao, J. (2026) · IEEE Transactions on Networking · doi:10.1109/ton.2026.3716703
“Given the excellent performance of LLMs in traditional natural language processing (NLP) tasks such as translation [7], summarization [8], and named entity recognition [9], as well as their unique contributions in network operations [10], [11], particularly in network configuration generation [12],…”
Dakuru, M., Lam, S. S., & Elluri, L. (2026) · Lecture notes in networks and systems · doi:10.1007/978-3-032-24360-7_26
Language-driven models for video anomaly detection: taxonomy, datasets, and future horizons
Saket, O., Aicha, A. B., & Fathallah, H. (2026) · International Journal of Multimedia Information Retrieval · doi:10.1007/s13735-026-00415-w
Aldashnan, B., Alothaim, A., & Alsanad, A. (2026) · ACM Transactions on Asian and Low-Resource Language Information Processing · doi:10.1145/3838729
“[63], the abstractiveness of summaries is measured through two key metrics: coverage and density.”
Enhancing LLM-Based Information Seeking with Search Trails: A Case Study
Garmendia, X., Díaz, Ó., & Winckler, M. A. A. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-22934-2_13
A Comparative Analysis of Transformer Models for Abstractive Summarization of Indian Legal Judgments
Mahalakshmi, G. S., Kumar, C. S., & Sendhilkumar, S. (2026) · Communications in computer and information science · doi:10.1007/978-3-032-35516-4_19
Jegede, O. O., & Senayon, E. (2026) · Universal Library of languages and literatures. · doi:10.70315/uloap.ullli.2026.0301004
Aftiss, A., Lamsiyah, S., Schommer, C., & Alaoui, S. O. E. (2026) · IEEE Access · doi:10.1109/access.2026.3729844
Contextual Learning Strategies for Dialogue Summarization Using Large Language Models
Penyasov, D. E., & Turalchuk, K. (2026) · Pattern Recognition and Image Analysis · doi:10.1134/s1054661826700252
The Media Bias Detector: A framework for annotating and analyzing the news
Haider, S., Tohidi, A., Wang, J. S., Dorr, T. G., Rothschild, D. M., Callison-Burch, C., & Watts, D. J. (2026) · Science Advances · doi:10.1126/sciadv.aea7456 · Cited by 1
“Prior work has found that LLMs perform well in generative tasks such as summarization ( 41 , 42 ), as well as discriminative tasks such as sentiment analysis and topic classification ( 43 , 44 ), where these tasks can even be performed in a zero-shot setting ( 45 ).”
LLM-generated scientific content detection method
Bader, A., & Alhijawi, B. (2026) · Neural Computing and Applications · doi:10.1007/s00521-026-12465-6
“For instance, LLMs can possess strong capabilities in news summarization [14 ], writing news articles that can be used to spread misinformation [15], creating short stories [16], generating restaurant reviews [17 ], and even answering exams such as the United States Medical Licensing Examination…”
Towards Argument Mining Agents: A Neuro-Symbolic Approach Based on Argumentation Schemes and LLMs
Trajano, G., Sarkadi, Ş., Gonçalves, A. L., & Panisson, A. R. (2025) · IEEE/WIC/ACM International Conference on Web Intelligence and Intelligent Agent Technology (WI-IAT) · doi:10.1109/wi-iat67162.2025.00055
“In parallel, several benchmarks have been introduced to evaluate these models across a range of NLP tasks, including code generation, semantic understanding, and mathematical reasoning [42], [43], [44].”
Maritime situational awareness: Harnessing big data and AI for textual risk mining and insights
Balasubramanian, L., Nguyen, T. T., & Deng, J. (2026) · Array · doi:10.1016/j.array.2026.101229
Shahabi, A. S., Kangavari, M. R., & Rahmani, A. M. (2026) · Information Processing & Management · doi:10.1016/j.ipm.2026.105152
Mussa, A., Туймебаев, Ж. К., Мансурова, М., & Shihab, A. H. (2026) · Information · doi:10.3390/info17090921
Evaluating Multimodal Input Combinations for Zero-Shot Summarization Across Indian LanguagesPreprint
Mazumder, S. (2026) · Research Square · doi:10.21203/rs.3.rs-11140524/v1
Costa, G., Mendes, R., Teixeira, B., & Ribeiro, L. (2026) · Journal of Computational and Data-Driven Materials Engineering · doi:10.68159/w258707955
Duarte-Martínez, V., Cobo, M. J., & Mendoza, J. N. (2026) · International Conference on eDemocracy & eGovernment · doi:10.1109/icedeg70169.2026.11695401
KOÇAK, B., & Yıldız, K. (2026) · Sakarya University Journal of Computer and Information Sciences · doi:10.35377/saucis...1864801
HuiBeom, K., Choi, D., & Kim, H. (2026) · Neurocomputing · doi:10.1016/j.neucom.2026.135275
Ristianto, R. P., Arisal, A., Kurniasari, D., & Khotimah, P. H. (2026) · International Conference on Computer, Control, Informatics and its Applications · doi:10.1109/ic3ina72033.2026.11714656
“Models trained to follow natural-language task descriptions summarize competently without task-specific fine-tuning, and benchmarking work reports that instruction-tuned systems generate coherent and informative news summaries in the zero-shot setting [2], [3], [4].”
Synthesizing Geographical Semantics and Large Language Model for OD ‐Based Trajectory Captioning
Zhou, H., Li, S., Ye, Y., Xu, Z., Wu, Y., Fan, C., Li, B., Yao, H. L., Liu, K., & Gong, J. (2026) · Transactions in GIS · doi:10.1111/tgis.70425
Qureshi, K. A., Sabih, M., & Siddiqui, M. S. (2026) · Future Internet · doi:10.3390/fi18100563
Su, S., Xu, J., Li, D., Chen, H., Zhang, X., Xia, L., & Pai, Z. (2026) · Journal of Manufacturing Systems · doi:10.1016/j.jmsy.2026.10.002
Metadata
- Authors
- Zhang, T., Ladhak, F., Durmus, E., Liang, P., McKeown, K., & Hashimoto, T. B.
- Affiliations
- Stanford University (US)
- Columbia University (US)
- Corresponding author
- Zhang, Tianyi
- Ladhak, Faisal
- Venue
- Transactions of the Association for Computational Linguistics
- Publisher
- MIT Press
- Published
- 2024
- Volume
- 12
- Pages
- 39-57
- ISSN
- 2307-387X (ISSN-L)
- OpenAlex
- W4391494845
- Cited by
- 415
- References
- 79
- Citations by year
- 2026: 108 · 2025: 165 · 2024: 96
- FWCI
- 54.46
- Type
- journal article
- Language
- en
- Open access
- diamond
- doi.org
- Licence
- Locations
- Transactions of the Association for Computational Linguistics (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
- arXiv (Cornell University) (cc-by)
- Funders
- Open Philanthropy Project
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Artificial Intelligence › Advanced Text Analysis Techniques
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Keywords
- large language models, news summarization, automatic summarization, human evaluation, instruction tuning, zero-shot summarization, fine-tuning, model size, human-written summaries, paraphrase, freelance writers
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.