Learning from Disagreement: A Survey
Open accessThe DOI matches its Crossref record; 23 of the 23 references it cites were found in a registry or an index.
Abstract
Many tasks in Natural Language Processing (NLP) and Computer Vision (CV) offer evidence that humans disagree, from objective tasks such as part-of-speech tagging to more subjective tasks such as classifying an image or deciding whether a proposition follows from certain premises. While most learning in artificial intelligence (AI) still relies on the assumption that a single (gold) interpretation exists for each item, a growing body of research aims to develop learning methods that do not rely on this assumption. In this survey, we review the evidence for disagreements on NLP and CV tasks, focusing on tasks for which substantial datasets containing this information have been created. We discuss the most popular approaches to training models from datasets containing multiple judgments potentially in disagreement. We systematically compare these different approaches by training them with each of the available datasets, considering several ways to evaluate the resulting models. Finally, we discuss the results in depth, focusing on four key research questions, and assess how the type of evaluation and the characteristics of a dataset determine the answers to these questions. Our results suggest, first of all, that even if we abandon the assumption of a gold standard, it is still essential to reach a consensus on how to evaluate models. This is because the relative performance of the various training methods is critically affected by the chosen form of evaluation. Secondly, we observed a strong dataset effect. With substantial datasets, providing many judgments by high-quality coders for each item, training directly with soft labels achieved better results than training from aggregated or even gold labels. This result holds for both hard and soft evaluation. But when the above conditions do not hold, leveraging both gold and soft labels generally achieved the best results in the hard evaluation. All datasets and models employed in this paper are freely available as supplementary materials.
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
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
Adam: A Method for Stochastic Optimization
Kingma, D. P., & Ba, J. (2014) · arXiv · doi:10.48550/arxiv.1412.6980 · Cited by 98,946
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
On Information and Sufficiency
Kullback, S., & Leibler, R. A. (1951) · The Annals of Mathematical Statistics · doi:10.1214/aoms/1177729694 · Cited by 20,193
Distilling the Knowledge in a Neural Network
Hinton, G., Vinyals, O., & Dean, J. (2015) · arXiv · doi:10.48550/arxiv.1503.02531 · Cited by 14,204
Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps
Simonyan, K., Vedaldi, A., & Zisserman, A. (2013) · arXiv · doi:10.48550/arxiv.1312.6034 · Cited by 5,370
LabelMe: A Database and Web-Based Tool for Image Annotation
Russell, B. C., Torralba, A., Murphy, K. P., & Freeman, W. T. (2008) · International Journal of Computer Vision · doi:10.1007/s11263-007-0090-8 · Cited by 4,210
Striving for Simplicity: The All Convolutional Net
Springenberg, J. T., Dosovitskiy, A., Brox, T., & Riedmiller, M. (2014) · arXiv · doi:10.48550/arxiv.1412.6806 · Cited by 3,447
A 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
Cheap and fast---but is it good?: evaluating non-expert annotations for natural language tasks
Snow, R., O'Connor, B., Jurafsky, D., & Ng, A. Y. (2008) · Proceedings of the Conference on Empirical Methods in Natural Language Processing - EMNLP '08 · doi:10.3115/1613715.1613751 · Cited by 1,887
Cited by
The “Problem” of Human Label Variation: On Ground Truth in Data, Modeling and Evaluation
Plank, B. (2022) · Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/2022.emnlp-main.731 · Cited by 121
Self-citationHashemi, H., Eisner, J., Rosset, C., Van Durme, B., & Kedzie, C. (2024) · Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2024.acl-long.745 · Cited by 47
Why Don’t You Do It Right? Analysing Annotators’ Disagreement in Subjective Tasks
Sandri, M., Leonardelli, E., Tonelli, S., & Jezek, E. (2023) · Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics · doi:10.18653/v1/2023.eacl-main.178 · Cited by 31
“Evaluation results can contribute to ongoing studies investigating the negative impact of disagreement on systems performance (Schwartz et al., 2011; Beigman Klebanov and Beigman, 2014; Leonardelli et al., 2021; Uma et al., 2021) by providing a more fine-grained view on disagreement types.”
BackgroundInfluentialCan Large Language Models Capture Dissenting Human Voices?
Lee, N., An, N. M., & Thorne, J. (2023) · arXiv · doi:10.48550/arxiv.2305.13788 · Cited by 22
The Value of Disagreement in AI Design, Evaluation, and Alignment
Fazelpour, S., & Fleisher, W. (2025) · Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency · doi:10.1145/3715275.3732146 · Cited by 10
“In dataset construction, annotators frequently diverge in their assessments due to differences in interpretation, cultural contexts, expertise, lived experiences, and more [63, 97, 117, 125].”
The Ecological Fallacy in Annotation: Modeling Human Label Variation goes beyond Sociodemographics
Orlikowski, M., Röttger, P., Cimiano, P., & Hovy, D. (2023) · Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) · doi:10.18653/v1/2023.acl-short.88 · Cited by 9
“These models are motivated by the concern that aggregating labels into a single “truth” is too simplistic for many tasks (Uma et al., 2021; Basile et al., 2021) and might introduce uneven representation of perspectives (Prabhakaran et al., 2021; Abercrombie et al., 2022).”
BackgroundSelf-citation
Metadata
- Authors
- Uma, A. N., Fornaciari, T., Hovy, D., Paun, S., Plank, B., & Poesio, M.
- Venue
- Journal of Artificial Intelligence Research
- Publisher
- AI Access Foundation
- Published
- 2021-12-27
- Volume
- 72
- Pages
- 1385-1470
- ISSN
- 1076-9757 (ISSN-L)
- OpenAlex
- W4200444669
- Cited by
- 144
- References
- 23
- FWCI
- 6.97
- 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.