Elamur Research⌕
☰
home / papers / 2026 / Fine-Grained Perspectives: Modeling Explanations with Annotator-Specific Rationales

Fine-Grained Perspectives: Modeling Explanations with Annotator-Specific Rationales

Paper · 2026 · Olufunke O. Sarumi, Charles Welch and Daniel Braun

Summary

Perspectivist NLP keeps individual annotators' labels instead of collapsing them into a majority vote. Yet the reasoning behind each annotator's choice is rarely modeled, even when datasets collect it. This paper treats annotator-written rationales as a fine-grained signal of perspective. It predicts each annotator's labels and generates that annotator's explanation together.

The authors use VariErrNLI, an NLI corpus with disaggregated labels and per-label explanations from several annotators. Their classifier is built on DeBERTa-v3-base and uses the User Passport method, which injects learned annotator embeddings and demographic metadata at the representation level. Two explainers sit on top of it:

  • a post-hoc explainer driven by prompts;
  • a "prefixed bridge" explainer, which feeds the classifier's annotator-conditioned representations into a generative model as a prefix.

Adding explanation modeling clearly improves label prediction over an annotator-aware classifier trained without it. With the bridge variant, predicted labels stay more consistently aligned and the explanations are more semantically coherent; the post-hoc variant, by contrast, yields explanations whose wording sits closer to the human ones. The work suggests that disagreement is captured more faithfully when models represent why annotators differ, not only what they chose.

Key points

  • Data: VariErrNLI (Weber-Genzel et al., 2024) contains about 500 MNLI items drawn from ChaosNLI, each annotated by four annotators.
    • Round 1 produced 1,933 label–explanation pairs. Labels are E, N or C, and annotators could assign more than one.
    • In Round 2, annotators judged whether each explanation plausibly supports its label.
    • The authors use the LeWiDi 2025 release, which has fixed train, dev and test splits.
  • Classifier: User Passport on a DeBERTa-v3-base encoder, combining annotator embeddings with metadata. A masked loss skips annotator–item pairs that have no annotation.
  • Explainers: a prompt-based post-hoc model and a prefixed bridge model that passes internal annotator-specific representations into a generator.
  • Findings: joint explanation modeling beats the annotator-aware classifier baseline. The bridge model leads on semantic consistency and label stability, and the post-hoc model leads on lexical overlap. Exact scores are not in the provided excerpt.
  • Code: github.com/Responsible-NLP/LRECNLPerspectives2026-Fine-Grained-Perspective

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Philipps University of MarburgMarburg · Germany8 works · 226 citations🎓McMaster UniversityHamilton · Canada4 works · 656 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

Cite

Cite as Olufunke O. Sarumi, Charles Welch and Daniel Braun. 2026. Fine-Grained Perspectives: Modeling Explanations with Annotator-Specific Rationales. Proceedings of the the fifth edition of NLPerspectives.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/4s7kqwvy5i6x

@inproceedings{sarumi2026fine,
  title = {{Fine-Grained Perspectives: Modeling Explanations with Annotator-Specific Rationales}},
  author = {Olufunke O. Sarumi, Charles Welch and Daniel Braun},
  booktitle = {Proceedings of the the fifth edition of NLPerspectives},
  year = {2026},
  doi = {10.63317/4s7kqwvy5i6x},
  url = {https://doi.org/10.63317/4s7kqwvy5i6x}
}

References

22 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Federico Cabitza, Andrea Campagner, and Valerio Basile. 2023. Toward a perspectivist turn in ground truthing for predictive computing. Proceedings of the AAAI Conference on Artificial Intelligence, 37(6):6860–6868.
  2. Lei Li, Yongfeng Zhang, and Li Chen. 2021b. Personalized transformer for explainable recommendation. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 49… find
  3. Aida Mostafazadeh Davani, Markos Markatou, Tommaso Fornaciari, Silviu Paun, Dirk Hovy, Joel Tetreault, and Cecilia Ovesdotter Alm. 2022. Dealing with disagreements: Looking beyond the majority vote in subjective annotations. Transactions of the Association for Computational Linguistics, 10:92–110.
  4. Matteo Mastromattei, Leonardo Ranaldi, Francesca Fallucchi, and Fabio Massimo Zanzotto. 2022a. Syntax and prejudice: ethically-charged biases of a syntax-based hate speech recognizer unveiled. PeerJ Computer Science, 8:e859. find
  5. Simona Frenda, Gavin Abercrombie, Valerio Basile, Alessandro Pedrani, Raffaella Panizzon, Alessandra Teresa Cignarella, Cristina Marco, and Davide Bernardi. 2025. Perspectivist approaches to natural language processing: a survey. Language Resources and Evaluation, 59:1719–1746. find
  6. Michele Mastromattei, Valerio Basile, and Fabio Massimo Zanzotto. 2022b. Change my mind: How syntax-based hate speech recognizer can uncover hidden motivations based on different viewpoints. In Proceedings of the 1st Workshop on Perspectivist Approaches to NLP @LREC2022, pages 117–125, Marseille, Fr… find
  7. Shirley Anugrah Hayati, Minhwa Lee, Dheeraj Rajagopal, and Dongyeop Kang. 2024. How far can we extract diverse perspectives from large language models? In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 5336–5366, Miami, Florida, USA. Association for Com… find
  8. Benedetta Muscato, Lucia Passaro, Gizem Gezici, and Fosca Giannotti. 2025. Perspectives in play: A multi-perspective approach for more inclusive nlp systems. In Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, IJCAI-2025, page 9827–9835. International Joint… find
  9. Ayushi Nirmal, Amrita Bhattacharjee, Paras Sheth, and Huan Liu. 2024. Towards interpretable hate speech detection using large language model-extracted rationales. In Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024), pages 223–233, Mexico City, Mexico. Association for Computation… find
  10. Charles Welch, Chenxi Gu, Jonathan K. Kummerfeld, Veronica Perez-Rosas, and Rada Mihalcea. 2022. Leveraging similar users for personalized language modeling with limited data. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 17… find
  11. Matthias Orlikowski, Jiaxin Pei, Paul Röttger, Philipp Cimiano, David Jurgens, and Dirk Hovy. 2025. Beyond demographics: Fine-tuning large language models to predict individuals’ subjective text perceptions.
  12. Jin Xu, Mariët Theune, and Daniel Braun. 2024. Leveraging annotator disagreement for text classification. In Proceedings of the 7th International Conference on Natural Language and Speech Processing (ICNLSP 2024), pages 1–10, Trento. Association for Computational Linguistics. find
  13. Ellie Pavlick and Tom Kwiatkowski. 2019. Inherent disagreements in human textual inferences. Transactions of the Association for Computational Linguistics, 7:677–694.
  14. Language Resource References Joan Plepi, Béla Neuendorf, Lucie Flek, and Charles Welch. 2022. Unifying data perspectivism and personalization: An application to social norms. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 7391–7402, Abu Dhabi, United… find
  15. Chung, Hyung Won and Hou, Le and Longpre, Shayne and Zoph, Barret and Tai, Yi and Fedus, William and Li, Yunxuan and Wang, Xuezhi and Dehghani, Mostafa and Brahma, Siddhartha and Webson, Albert and Gu, Shixiang Shane and Dai, Zhuyun and Suzgun, Mirac and Chen, Xinyun and Chowdhery, Aakanksha and Cas… find
  16. Joan Plepi, Charles Welch, and Lucie Flek. 2024. Perspective taking through generating responses to conflict situations. In Findings of the Association for Computational Linguistics: ACL 2024, pages 6482–6497, Bangkok, Thailand. Association for Computational Linguistics. find
  17. Olufunke O. Sarumi, Béla Neuendorf, Joan Plepi, Lucie Flek, Jörg Schlötterer, and Charles Welch. 2024. Corpus considerations for annotator modeling and scaling. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Techno… find
  18. Pengcheng He and Jianfeng Gao and Weizhu Chen. 2023. DeBERTaV3: Improving De-BERTa using ELECTRA-Style Pre-Training with Gradient-Disentangled Embedding Sharing. find
  19. Elisa Leonardelli and Silvia Casola and Siyao Peng and Giulia Rizzi and Valerio Basile and Elisabetta Fersini and Diego Frassinelli and Hyewon Jang and Maja Pavlovic and Barbara Plank and Massimo Poesio. 2026. LeWiDi-2025 at NLPerspectives: Third Edition of the Learning with Disagreements Shared Tas… find
  20. Olufunke O. Sarumi, Charles Welch, and Daniel Braun. 2025. NLP-ResTeam at LeWiDi-2025:performance shifts in perspective aware models based on evaluation metrics. In Proceedings of the The 4th Workshop on Perspectivist Approaches to NLP, pages 219–227, Suzhou, China. Association for Computational Lin… find
  21. Weber-Genzel, Leon and Peng, Siyao and De Marneffe, Marie-Catherine and Plank, Barbara. 2024. VariErr NLI: Separating Annotation Error from Human Label Variation. Association for Computational Linguistics.
  22. Alexandra Uma, Tommaso Fornaciari, Anca Dumitrache, Tristan Miller, Jon Chamberlain, Barbara Plank, Edwin Simpson, and Massimo Poesio. 2021. SemEval-2021 task 12: Learning with disagreements. In Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021), pages 338–347, Onli… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sarumi-2026-fine-grained-perspectives-modeling-explanations/