Ranking Generated Summaries by Correctness: An Interesting but Challenging Application for Natural Language Inference
Open accessThe DOI matches its Crossref record; 30 of the 30 references it cites were found in a registry or an index.
Abstract
While recent progress on abstractive summarization has led to remarkably fluent summaries, factual errors in generated summaries still severely limit their use in practice.In this paper, we evaluate summaries produced by state-of-the-art models via crowdsourcing and show that such errors occur frequently, in particular with more abstractive models.We study whether textual entailment predictions can be used to detect such errors and if they can be reduced by reranking alternative predicted summaries.That leads to an interesting downstream application for entailment models.In our experiments, we find that outof-the-box entailment models trained on NLI datasets do not yet offer the desired performance for the downstream task and we therefore release our annotations as additional test data for future extrinsic evaluations of NLI.
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
“Textual entailment has previously been shown to correlate with faithfulness10 in the context of factual consistency29 as well as being used to measure factuality in abstractive summarization30, especially when applied at the right granularity31.”
From Detecting hallucinations in large language models using semantic entropy (2024) · cites this paper as [30]
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
“It has been shown that state-of-the-art abstractive summarization systems, evaluated with metrics such as ROUGE, BLEU, and METEOR, have hallucinated content in 25% of their generated summaries [66].”
From Survey of Hallucination in Natural Language Generation (2023) · cites this paper as [66]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Nevertheless, as illustrated by Falke et al. [66], off-the-shelf NLI models tend to transfer poorly to the abstractive summarization task.”
From Survey of Hallucination in Natural Language Generation (2023) · cites this paper as [66]
The cited source says what the claim attributes to it.
In the citing paper
“In line with this, numerous studies [82, 208] have trained classifiers on NLI datasets to identify factual inaccuracies, especially in the context of abstract summarization.”
From A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions (2025) · cites this paper as [82]
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
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2018) · arXiv · doi:10.48550/arxiv.1810.04805 · Cited by 73,173 · cited 1×
“And finally, we include a version of BERT-base (Devlin et al., 2019) fine-tuned on MultiNLI (Williams et al., 2018 crowdsourced correctness labels for the resulting 1000 summaries.”
MethodBERT: 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 1×
“And finally, we include a version of BERT-base (Devlin et al., 2019) fine-tuned on MultiNLI (Williams et al., 2018 crowdsourced correctness labels for the resulting 1000 summaries.”
MethodDeep Contextualized Word Representations
Peters, M., Neumann, M., Iyyer, M., Gardner, M., Clark, C., Lee, K., & Zettlemoyer, L. (2018) · Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers) · doi:10.18653/v1/n18-1202 · Cited by 11,983 · cited 1×
“(Peters et al., 2018) by Al-lenNLP.”
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×
“PGC (See et al., 2017) 8% 39.”
MethodA large annotated corpus for learning natural language inference
Bowman, S. R., Angeli, G., Potts, C., & Manning, C. D. (2015) · Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d15-1075 · Cited by 3,377 · cited 3×
“We study the use of natural language inference (NLI) (Bowman et al., 2015), also known as textual entailment (Dagan et al., 2006), to detect factual errors.”
BackgroundA Broad-Coverage Challenge Corpus for Sentence Understanding through Inference
Williams, A., Nangia, N., & Bowman, S. (2018) · Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers) · doi:10.18653/v1/n18-1101 · Cited by 3,342 · cited 2×
“And finally, we include a version of BERT-base (Devlin et al., 2019) fine-tuned on MultiNLI (Williams et al., 2018 crowdsourced correctness labels for the resulting 1000 summaries.”
MethodA Neural Attention Model for Abstractive Sentence Summarization
Rush, A. M., Chopra, S., & Weston, J. (2015) · Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d15-1044 · Cited by 2,656 · cited 1×
“Following the pioneering works of Rush et al. (2015), Chopra et al. (2016) and Nallapati et al. (2016), many models have been developed in recent years that can all generate summaries by freely choosing words from a large vocabulary rather than reusing full sentences from the input document.”
BackgroundInfluentialTeaching Machines to Read and Comprehend
Hermann, K. M., Kočiský, T., Grefenstette, E., Espeholt, L., Kay, W., Suleyman, M., & Blunsom, P. (2015) · arXiv (Cornell University) · doi:10.48550/arxiv.1506.03340 · Cited by 2,521 · cited 1×
“The general success of deep learning techniques and the availability of large-scale single-document summarization datasets, such as the CNN-DailyMail (CNN-DM) corpus (Hermann et al., 2015), have recently led to a renewed interest in abstractive summarization.”
BackgroundAbstractive Text Summarization using Sequence-to-sequence RNNs and Beyond
Nallapati, R., Zhou, B., dos Santos, C., Gulcehre, C., & Xiang, B. (2016) · Proceedings of The 20th SIGNLL Conference on Computational Natural Language Learning · doi:10.18653/v1/k16-1028 · Cited by 2,361 · cited 1×
“Following the pioneering works of Rush et al. (2015), Chopra et al. (2016) and Nallapati et al. (2016), many models have been developed in recent years that can all generate summaries by freely choosing words from a large vocabulary rather than reusing full sentences from the input document.”
BackgroundInfluentialSupervised Learning of Universal Sentence Representations from Natural Language Inference Data
Conneau, A., Kiela, D., Schwenk, H., Barrault, L., & Bordes, A. (2017) · Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d17-1070 · Cited by 2,049 · cited 3×
“4 Further, we also include our own implementations of InferSent (Conneau et al., 2017) and shortcut-stacked encoders (SSE) (Nie and Bansal).”
MethodInfluentialThe PASCAL Recognising Textual Entailment Challenge
Dagan, I., Glickman, O., & Magnini, B. (2006) · Lecture Notes in Computer Science · doi:10.1007/11736790_9 · Cited by 1,628
Self-citationA Decomposable Attention Model for Natural Language Inference
Parikh, A., Täckström, O., Das, D., & Uszkoreit, J. (2016) · Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d16-1244 · Cited by 1,296
Enhanced LSTM for Natural Language Inference
Chen, Q., Zhu, X., Ling, Z.-H., Wei, S., Jiang, H., & Inkpen, D. (2017) · Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/p17-1152 · Cited by 1,113
Right for the Wrong Reasons: Diagnosing Syntactic Heuristics in Natural Language Inference
McCoy, R. T., Pavlick, E., & Linzen, T. (2019) · Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/p19-1334 · Cited by 944 · cited 1×
“Another known issue is that NLI models tend to rely on simplifying heuristics such as lexical overlap (McCoy et al., 2019 current NLI models are not yet robust enough for our downstream task.”
BackgroundAbstractive Sentence Summarization with Attentive Recurrent Neural Networks
Chopra, S., Auli, M., & Rush, A. M. (2016) · Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies · doi:10.18653/v1/n16-1012 · Cited by 881 · cited 1×
“Following the pioneering works of Rush et al. (2015), Chopra et al. (2016) and Nallapati et al. (2016), many models have been developed in recent years that can all generate summaries by freely choosing words from a large vocabulary rather than reusing full sentences from the input document.”
BackgroundInfluentialBottom-Up Abstractive Summarization
Gehrmann, S., Deng, Y., & Rush, A. (2018) · Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d18-1443 · Cited by 667 · cited 1×
“BUS The bottom-up summarization system recently proposed by Gehrmann et al. (2018).”
Fast Abstractive Summarization with Reinforce-Selected Sentence Rewriting
Chen, Y.-C., & Bansal, M. (2018) · Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/p18-1063 · Cited by 595 · cited 2×
“Incorrect FAS The hybrid extractive-abstractive system proposed by Chen and Bansal (2018) including their redundancy-based reranking.”
MethodLearning Whom to Trust with MACE
Hovy, D., Berg-Kirkpatrick, T., Vaswani, A., & Hovy, E. H. (2013) · Archivio istituzionale della Ricerca - Bocconi · Cited by 504
Faithful to the Original: Fact Aware Neural Abstractive Summarization
Cao, Z., Wei, F., Li, W., & Li, S. (2018) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v32i1.11912 · Cited by 355
A Simple, Fast Diverse Decoding Algorithm for Neural Generation
Li, J., Monroe, W., & Jurafsky, D. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1611.08562 · Cited by 195 · cited 1×
“Since the availability of at least one correct summary hypothesis is a prerequisite of the reranking approach, we rely on FAS which uses a variant of beam search yielding more diverse hypotheses (Li et al., 2016).”
MethodFaithful to the Original: Fact Aware Neural Abstractive SummarizationPreprint
Cao, Z., Wei, F., Li, W., et al. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1711.04434 · Cited by 173 · cited 5×
“Similar to previous work by Cao et al. (2018) and Li et al. (2018), we argue that the correctness of a generated summary can only be reliably evaluated by manual inspection.”
BackgroundMethodInfluentialMulti-Reward Reinforced Summarization with Saliency and Entailment
Pasunuru, R., & Bansal, M. (2018) · Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers) · doi:10.18653/v1/n18-2102 · Cited by 162 · cited 1×
“Moreover, Guo et al. (2018) and Pasunuru and Bansal (2018) proposed to use NLI-based loss functions or multi-task learning with NLI for document summarization.”
MethodLi, H., Zhu, J., Zhang, J., & Zong, C. (2018) · International Conference on Computational Linguistics · Cited by 134
Soft Layer-Specific Multi-Task Summarization with Entailment and Question Generation
Guo, H., Pasunuru, R., & Bansal, M. (2018) · Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/p18-1064 · Cited by 129 · cited 1×
“Moreover, Guo et al. (2018) and Pasunuru and Bansal (2018) proposed to use NLI-based loss functions or multi-task learning with NLI for document summarization.”
MethodCorpus Annotation through Crowdsourcing: Towards Best Practice Guidelines
Sabou, M., Bontcheva, K., Derczynski, L., & Arno Scharl, A. (2014) · Language Resources and Evaluation Conference (LREC) · doi:10.63317/569wpuc5pjnt · Cited by 128 · cited 1×
“A challenge of crowdsourcing is that workers are untrained and some might produce low quality annotations (Sabou et al., 2014).”
BackgroundShortcut-Stacked Sentence Encoders for Multi-Domain Inference
Nie, Y., & Bansal, M. (2017) · Proceedings of the 2nd Workshop on Evaluating Vector Space Representations for NLP · doi:10.18653/v1/w17-5308 · Cited by 127
Human vs. Muppet: A Conservative Estimate of Human Performance on the GLUE Benchmark
Nangia, N., & Bowman, S. R. (2019) · Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/p19-1449 · Cited by 48 · cited 1×
“On the other hand, the state-of-the-art performance on common NLI datasets is already very close to human performance (Nikita and Bowman, 2019), suggesting that new datasets, such as the one presented here, are necessary to expose the models’ remaining limitations.”
BackgroundShortcut-Stacked Sentence Encoders for Multi-Domain InferencePreprint
Nie, Y., & Bansal, M. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1708.02312 · Cited by 37
Towards Improving Abstractive Summarization via Entailment Generation
Pasunuru, R., Guo, H., & Bansal, M. (2017) · Proceedings of the Workshop on New Frontiers in Summarization · doi:10.18653/v1/w17-4504 · Cited by 33 · cited 1×
“multi-task learning with NLI (Li et al., 2018; Pasunuru et al., 2017) to successfully improve the correctness of abstractive sentence summaries.”
Robust Neural Abstractive Summarization Systems and Evaluation against Adversarial Information
Fan, L., Yu, D., & Wang, L. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1810.06065 · Cited by 18 · cited 1×
“Very recently, Fan et al. (2018) showed that with ideas similar to Cao et al. (2018)’s work, the correctness of document summaries can also be improved.”
Background
Research works printed but not linked: 4
- [1]1. Samuel R. Bowman, Gabor Angeli, Christopher Potts, and Christopher D. Manning. 2015. A Large An- notated Corpus for Learning Natural Language In- ference. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Process- ing, pages 632-642, Lisbon, Portugal.
- [4]4. Yen-Chun Chen and Mohit Bansal. 2018. Fast Abstrac- tive Summarization with Reinforce-Selected Sen- tence Rewriting. In Proceedings of the 56th Annual Meeting of the Association for Computational Lin- guistics, pages 675-686, Melbourne, Australia.
- [10]10. Sebastian Gehrmann, Yuntian Deng, and Alexander Rush. 2018. Bottom-Up Abstractive Summariza- tion. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Process- ing, pages 4098-4109, Brussels, Belgium.
- [23]23. Matthew Peters, Mark Neumann, Mohit Iyyer, Matt Gardner, Christopher Clark, Kenton Lee, and Luke Zettlemoyer. 2018. Deep Contextualized Word Rep- resentations. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 2227-2237, New Orleans, LA, USA.
Cited by
Survey of Hallucination in Natural Language Generation
Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y. J., Madotto, A., & Fung, P. (2023) · ACM Computing Surveys · doi:10.1145/3571730 · Cited by 5,025
“[45] discover that models trained on NLI datasets can not transfer well to abstractive summarization tasks, degrading the reliability of NLI-based hallucination metrics.”
BackgroundInfluentialSurvey of Hallucination in Natural Language Generation
Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y., Chen, D., Dai, W., et al. (2022) · arXiv · doi:10.48550/arxiv.2202.03629 · Cited by 4,981
Huang, L., Yu, W., Ma, W., Zhong, W., Feng, Z., Wang, H., Chen, Q., Peng, W., Feng, X., Qin, B., et al. (2025) · ACM Transactions on Information Systems · doi:10.1145/3703155 · Cited by 4,688
“In line with this, numerous studies (Falke et al., 2019; Maynez et al., 2020) have trained classifiers on NLI datasets to identify factual inaccuracies, especially in the context of abstract summarization.”
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
“In line with our findings, Falke et al. (2019) observed that the BERT-based NLI models substantially improved summaries reranking in terms of their correctness.”
BackgroundResult
Metadata
- Authors
- Falke, T., Ribeiro, L. F. R., Utama, P. A., Dagan, I., & Gurevych, I.
- Affiliations
- Amazon (United States) (US)
- Technische Universität Darmstadt (DE)
- Bar-Ilan University (IL)
- Venue
- Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics
- Published
- 2019
- Pages
- 2214-2220
- ISSN
- 0736-587X (ISSN-L)
- OpenAlex
- W2951211142
- Semantic Scholar
- 470c7c26f16bc52ea68a159c84f07b78d58fd02a
- Cited by
- 265
- References
- 28
- Citations by year
- 2026: 11 · 2025: 16 · 2024: 19 · 2023: 68 · 2022: 60 · 2021: 57 · 2020: 28 · 2019: 6
- FWCI
- 14.45
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- Locations
- Proceedings of the conference - Association for Computational Linguistics. Meeting (publishedVersion, cc-by)
- TUbilio (Technical University of Darmstadt) (acceptedVersion)
- TUbilio (Technical University of Darmstadt) (submittedVersion)
- Funders
- Deutsche Forschungsgemeinschaft (GRK 1994, GRK 1994/1, DA 1600/1-1, GU 798/17-1)
- Technische Universität Darmstadt
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Physical Sciences › Computer Science › Artificial Intelligence › Advanced Text Analysis Techniques
- Keywords
- abstractive summarization, factual errors, natural language inference, crowdsourced evaluation, external evaluation
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.