Dealing with Disagreements: Looking Beyond the Majority Vote in Subjective Annotations
Open accessThe DOI matches its Crossref record; 26 of the 26 references it cites were found in a registry or an index.
Abstract
Majority voting and averaging are common approaches used to resolve annotator disagreements and derive single ground truth labels from multiple annotations. However, annotators may systematically disagree with one another, often reflecting their individual biases and values, especially in the case of subjective tasks such as detecting affect, aggression, and hate speech. Annotator disagreements may capture important nuances in such tasks that are often ignored while aggregating annotations to a single ground truth. In order to address this, we investigate the efficacy of multi-annotator models. In particular, our multi-task based approach treats predicting each annotators’ judgements as separate subtasks, while sharing a common learned representation of the task. We show that this approach yields same or better performance than aggregating labels in the data prior to training across seven different binary classification tasks. Our approach also provides a way to estimate uncertainty in predictions, which we demonstrate better correlate with annotation disagreements than traditional methods. Being able to model uncertainty is especially useful in deployment scenarios where knowing when not to make a prediction is important.
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
BERT: 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
An argument for basic emotions
Ekman, P. (1992) · Cognition and Emotion · doi:10.1080/02699939208411068 · Cited by 9,646 · cited 3×
“While several datasets (Strapparava and Mihalcea, 2007; Buechel and Hahn, 2017) include a small set of labels representing the six Ekman emotions (Ek-man, 1992) — anger , disgust , fear , joy , sadness , and surprise ), or bipolar dimensions of affect — arousal and valence (Russell, 2003), others such as Demszky et al…”
MethodTransformers: State-of-the-Art Natural Language Processing
Wolf, T., Debut, L., Sanh, V., Chaumond, J., Delangue, C., Moi, A., Cistac, P., Rault, T., Louf, R., Funtowicz, M., et al. (2020) · Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations · doi:10.18653/v1/2020.emnlp-demos.6 · Cited by 8,741
Dropout as a bayesian approximation: Representing model uncertainty in deep learning
Gal, Y., & Ghahramani, Z. (2015) · arXiv · doi:10.48550/arxiv.1506.02142 · Cited by 6,471
Core affect and the psychological construction of emotion.
Russell, J. A. (2003) · Psychological Review · doi:10.1037/0033-295x.110.1.145 · Cited by 6,113 · cited 2×
“…the six Ekman emotions (Ek-man, 1992) — anger , disgust , fear , joy , sadness , and surprise ), or bipolar dimensions of affect — arousal and valence (Russell, 2003), others such as Demszky et al. (2020) and Crowdflower (2016) include a wider range of emotion labels according to the Plutchik…”
BackgroundAutomated Hate Speech Detection and the Problem of Offensive Language
Davidson, T., Warmsley, D., Macy, M., & Weber, I. (2017) · Proceedings of the International AAAI Conference on Web and Social Media · doi:10.1609/icwsm.v11i1.14955 · Cited by 2,610
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
Advances in natural language processing
Hirschberg, J., & Manning, C. D. (2015) · Science · doi:10.1126/science.aaa8685 · Cited by 1,712 · cited 1×
“Detecting emotions from language has been a significant area of research in NLP for the past two decades (Liscombe et al., 2003; Aman and Szpakowicz, 2007; Desmet and Hoste, 2013; Hirschberg and Manning, 2015; Poria et al., 2019).”
BackgroundMaximum Likelihood Estimation of Observer Error-Rates Using the EM Algorithm
Dawid, A. P., & Skene, A. M. (1979) · Applied Statistics · doi:10.2307/2346806 · Cited by 1,692 · cited 1×
“In early work, Dawid and Skene (1979) used the EM algorithm to obtain maximum likelihood estimates of the “true” label to account for annotator errors.”
MethodGoEmotions: A Dataset of Fine-Grained Emotions
Demszky, D., Movshovitz-Attias, D., Ko, J., Cowen, A., Nemade, G., & Ravi, S. (2020) · Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/2020.acl-main.372 · Cited by 873
Cited by
Bias and Fairness in Large Language Models: A Survey
Gallegos, I. O., Rossi, R. A., Barrow, J., Tanjim, M. M., Kim, S., Dernoncourt, F., Yu, T., Zhang, R., & Ahmed, N. K. (2024) · Computational Linguistics · doi:10.1162/coli_a_00524 · Cited by 1,245
ToxiGen: A Large-Scale Machine-Generated Dataset for Adversarial and Implicit Hate Speech Detection
Hartvigsen, T., Gabriel, S., Palangi, H., Sap, M., Ray, D., & Kamar, E. (2022) · Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2022.acl-long.234 · Cited by 279
ToxiGen: A Large-Scale Machine-Generated Dataset for Adversarial and Implicit Hate Speech Detection
Hartvigsen, T., Gabriel, S., Palangi, H., Sap, M., Ray, D., & Kamar, E. (2022) · arXiv · doi:10.48550/arxiv.2203.09509 · Cited by 276
Who Validates the Validators? Aligning LLM-Assisted Evaluation of LLM Outputs with Human Preferences
Shankar, S., Zamfirescu-Pereira, J. D., Hartmann, B., Parameswaran, A., & Arawjo, I. (2024) · Proceedings of the 37th Annual ACM Symposium on User Interface Software and Technology · doi:10.1145/3654777.3676450 · Cited by 151
Feng, S., Park, C. Y., Liu, Y., & Tsvetkov, Y. (2023) · Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2023.acl-long.656 · Cited by 149
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
Algorithmic inclusion: Shaping the predictive algorithms of artificial intelligence in hiring
Kelan, E. K. (2024) · Human Resource Management Journal · doi:10.1111/1748-8583.12511 · Cited by 106
Challenges and Strategies in Cross-Cultural NLP
Hershcovich, D., Frank, S., Lent, H., de Lhoneux, M., Abdou, M., Brandl, S., Bugliarello, E., Cabello Piqueras, L., Chalkidis, I., Cui, R., et al. (2022) · Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2022.acl-long.482 · Cited by 99
Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback
Casper, S., Davies, X., Shi, C., Gilbert, T. K., Scheurer, J., Rando, J., Freedman, R., Korbak, T., Lindner, D., Freire, P., et al. (2023) · arXiv · doi:10.48550/arxiv.2307.15217 · Cited by 101
“The majority-rule problem with RLHF can also be improved by algorithms that explicitly model multiple evaluators [Gordon et al., 2021, Davani et al., 2022, Daniels-Koch and Freedman, 2022, Gordon et al., 2022, Barnett et al., 2023], that tune models to individuals [Kumar et al., 2021], or that use…”
BackgroundFarsight: Fostering Responsible AI Awareness During AI Application Prototyping
Wang, Z. J., Kulkarni, C., Wilcox, L., Terry, M., & Madaio, M. (2024) · Proceedings of the CHI Conference on Human Factors in Computing Systems · doi:10.1145/3613904.3642335 · Cited by 55
A hunt for the Snark: Annotator Diversity in Data Practices
Kapania, S., Taylor, A. S., & Wang, D. (2023) · Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3544548.3580645 · Cited by 50
“Researchers have turned to disagreement as a signal for deepening their understanding of the task and the data [2, 18, 76].”
BackgroundMethodWhy 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
From Bias to Repair: Error as a Site of Collaboration and Negotiation in Applied Data Science Work
Lin, C. K., & Jackson, S. J. (2023) · Proceedings of the ACM on Human-Computer Interaction · doi:10.1145/3579607 · Cited by 22
“Davani et al. have embraced both disagreement between and inconsistency within annotators because they argue that such variance provides more “ flexibility ” and “ better estimates for uncertainty ” to model the real-life behavior of annotations [25].”
BackgroundFairPrism: Evaluating Fairness-Related Harms in Text Generation
Fleisig, E., Amstutz, A., Atalla, C., Blodgett, S. L., Daumé III, H., Olteanu, A., Sheng, E., Vann, D., & Wallach, H. (2023) · Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2023.acl-long.343 · Cited by 14
“…judgments (on a scale) and then use the variance in the predicted scores to determine whether the classifier has low confidence or whether there is genuine disagreement in the real world (Palomaki et al., 2018; Pavlick and Kwiatkowski, 2019; Prabhakaran et al., 2021; Davani et al., 2022).”
BackgroundThe 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
“These approaches have been promising, resulting in epistemically and ethically improved systems [25, 75, 84].”
InfluentialThe 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
A framework for mitigating malicious RLHF feedback in LLM training using consensus based reward
Haider, Z., Rahman, M. H., Devabhaktuni, V., Moeykens, S., & Chakraborty, P. (2025) · Scientific Reports · doi:10.1038/s41598-025-92889-7 · Cited by 9
The Role of Expertise in Effectively Moderating Harmful Social Media Content
Abdelkadir, N. A., Yang, T., Kapania, S., Estefanos, M., Gebrekidan, F. B., Zelalem, Z., Ali, M., Berhe, R., Baker, D., Talat, Z., et al. (2025) · Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3706598.3714010 · Cited by 5
NLPositionality: Characterizing Design Biases of Datasets and Models
Santy, S., Liang, J. T., Bras, R. L., Reinecke, K., & Sap, M. (2023) · arXiv · doi:10.48550/arxiv.2306.01943 · Cited by 31
“This allows our framework to continuously collect new annotations and reflect more up-to-date measurements of design biases for free over long periods of time, compared to one-time paid studies such as in previous works (Sap et al., 2022; Davani et al., 2022).”
BackgroundResultTang, X., Li, J., & Wu, S. (2026) · Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3772318.3790405 · Cited by 4
“It calls researchers to be sensitive to practices erasing the diversity of human perception, such as rigid organizational control [1] and majority voting [24].”
NUTMEG: Separating Signal From Noise in Annotator Disagreement
Ivey, J., Gauch, S., & Jurgens, D. (2025) · Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/2025.emnlp-main.144
From Noise to Nuance: Enriching Subjective Data Annotation through Qualitative Analysis
Wan, R., Wang, H., Huang, T.-H. K., & Gao, J. (2025) · Proceedings of the Fourth Workshop on Bridging Human-Computer Interaction and Natural Language Processing (HCI+NLP) · doi:10.18653/v1/2025.hcinlp-1.20
Metadata
- Authors
- Davani, A. M., Díaz, M., & Prabhakaran, V.
- Venue
- Transactions of the Association for Computational Linguistics
- Publisher
- MIT Press
- Published
- 2022
- Volume
- 10
- Pages
- 92-110
- ISSN
- 2307-387X (ISSN-L)
- OpenAlex
- W3206428286
- Cited by
- 219
- References
- 26
- FWCI
- 14.30
- 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.