Elamur Research⌕
☰
home / papers / 2026 / Who Plays Which Role? Protagonist Detection and Classification in Moral Discourse

Who Plays Which Role? Protagonist Detection and Classification in Moral Discourse

Paper · 2026 · Mirko Sommer and Maria Becker

Summary

Computational work on moral language has mostly focused on detecting moral values and has largely ignored the actors who make, receive, or are affected by moral demands. This paper models these protagonists at the phrase level in the German Moralization Corpus (MorCorp; Becker et al., 2025), which covers political debates, news, and online discussions. Earlier feasibility work handled protagonists inside one unified chain-of-thought prompting setup. The authors instead split the problem into two stages. First, they locate protagonist spans. Second, they assign each span an actor group (individual, generic, institution/organization, social group) and one or more argumentative roles (demander, addressee, beneficiary, maleficiary).

They compare fine-tuned lightweight models, adapted NER models and prompted LLMs. They also set human baselines and run an annotation experiment on how context affects labeling. Fine-tuned NER models detect protagonists about as well as LLMs at far lower cost. Role labeling benefits most from contextual prompting, and the strongest systems reach or exceed human performance. Humans use context in more linguistically coherent ways than LLMs do. The study shows that decomposing the task makes analyses of moral responsibility more accurate and easier to interpret.

Key points

  • Data: MorCorp contains 11,503 German paragraphs of one to five sentences. Only the moralizing ones (about 18%) are used, with the original 70:15:15 split balanced by genre.
  • Label distribution: institutions make up 32% of protagonists, social groups 30%, individuals 20% and generic references 15%. Per instance, there are 0.65 beneficiaries, 0.64 addressees, 0.42 demanders and 0.10 maleficiaries. Role labels can overlap; for example, an inclusive "we" can be both demander and addressee.
  • Prior baseline: the earlier all-in-one prompting approach peaked at F1 0.19 for groups and 0.18 for roles (gpt-5-mini, few-shot). The decomposed pipeline outperforms it.
  • Context analysis: when given context, humans revise labels selectively based on syntactic and semantic cues. LLMs behave less consistently; for instance, 28% of Claude's role changes involve generic expressions.
  • Limitations and resources: the study covers German only, fine-tunes only small to mid-sized models, and faces inherently ambiguous role annotation. Code, predictions and supplementary material are on GitHub (GS-Uni-Heidelberg/Paper-WhoPlaysWhichRole).

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

🎓Heidelberg UniversityHeidelberg · Germany38 works · 41,356 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

🧭 Topics 1Research subjects and languages the work is about.Discourse analysis · 466 works

Cite

Cite as Mirko Sommer and Maria Becker. 2026. Who Plays Which Role? Protagonist Detection and Classification in Moral Discourse. Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 4: Student Research Workshop).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.eacl-srw.27

@inproceedings{sommer2026who,
  title = {{Who Plays Which Role? Protagonist Detection and Classification in Moral Discourse}},
  author = {Mirko Sommer and Maria Becker},
  booktitle = {Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 4: Student Research Workshop)},
  year = {2026},
  doi = {10.18653/v1/2026.eacl-srw.27},
  url = {https://doi.org/10.18653/v1/2026.eacl-srw.27}
}

References

31 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Alan Akbik, Duncan Blythe, and Roland Vollgraf. 2018. Contextual string embeddings for sequence labeling. In COLING 2018, 27th International Conference on Computational Linguistics, pages 1638–1649.
  2. Maria Becker, Mirko Sommer, Lars Tapken, Yi Wan Teh, and Bruno Brocai. 2025. The moralization corpus: Frame-based annotation and analysis of moralizing speech acts across diverse text genres. Preprint, arXiv:2512.15248. find
  3. Tobias Bornheim, Niklas Grieger, Patrick Gustav Blaneck, and Stephan Bialonski. 2024. Speaker attribution in German parliamentary debates with QLoRA-adapted large language models. Journal for Language Technology and Computational Linguistics, 37(1):1–13. find
  4. Luana Bulla, Stefano De Giorgis, Misael Mongiovì, and Aldo Gangemi. 2025. Large language models meet moral values: A comprehensive assessment of moral abilities. Computers in Human Behavior Reports, 17:100609. find
  5. Luana Bulla, Aldo Gangemi, and Misael Mongiovì. 2024. Do language models understand morality? towards a robust detection of moral content. In Value Engineering in Artificial Intelligence, pages 98–113, Cham. Springer Nature Switzerland. find
  6. Ted Byrt, Janet Bishop, and John B Carlin. 1993. Bias, prevalence and kappa. Journal of clinical epidemiology, 46(5):423–429. find
  7. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. Preprint, arXiv:1810.04805.
  8. Thomas G Dietterich. 1998. Approximate statistical tests for comparing supervised classification learning algorithms. Neural computation, 10(7):1895–1923. find
  9. Neele Falk and Gabriella Lapesa. 2025. Mining the uncertainty patterns of humans and models in the annotation of moral foundations and human values. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 22898–22921. find
  10. Ines Reinig, Maria Becker, Ines Rehbein, and Simone Ponzetto. 2024. A survey on modelling morality for text analysis. In Findings of the Association for Computational Linguistics: ACL 2024, pages 4136–4155, Bangkok, Thailand. Association for Computational Linguistics. find
  11. Sayontan Ghosh, Mahnaz Koupaee, Isabella Chen, Francis Ferraro, Nathanael Chambers, and Niranjan Balasubramanian. 2023. PASTA: A dataset for modeling PArticipant STAtes in narratives. Transactions of the Association for Computational Linguistics, 11:1283– 1300. find
  12. Michael Roth and Mirella Lapata. 2015. Context-aware frame-semantic role labeling. Transactions of the Association for Computational Linguistics, 3:449– 460.
  13. Jesse Graham, Jonathan Haidt, Sena Koleva, Matt Motyl, Ravi Iyer, Sean P. Wojcik, and Peter H. Ditto. 2013. Chapter two - moral foundations theory: The pragmatic validity of moral pluralism. In Patricia Devine and Ashby Plant, editors, Advances in Experimental Social Psychology, volume 47, pages 55–… find
  14. Jonathan Haidt and Jesse Graham. 2007. When Morality Opposes Justice: Conservatives Have Moral Intuitions that Liberals may not Recognize. Social Justice Research, 20(1):98–116. find
  15. Jonathan Haidt and Craig Joseph. 2004. Intuitive ethics: How innately prepared intuitions generate culturally variable virtues. Daedalus, 133(4):55–66. find
  16. Shamik Roy, Nishanth Sridhar Nakshatri, and Dan Goldwasser. 2022. Towards few-shot identification of morality frames using in-context learning. In Proceedings of the Fifth Workshop on Natural Language Processing and Computational Social Science (NLP+CSS), pages 183–196, Abu Dhabi, UAE. Association f… find
  17. Shamik Roy, Maria Leonor Pacheco, and Dan Goldwasser. 2021. Identifying morality frames in political tweets using relational learning. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 9939–9958, Online and Punta Cana, Dominican Republic. Association fo… find
  18. Josef Ruppenhofer, Caroline Sporleder, Roser Morante, Collin Baker, and Martha Palmer. 2009. SemEval-2010 task 10: Linking events and their participants in discourse. In Proceedings of the Workshop on Semantic Evaluations: Recent Achievements and Future Directions (SEW-2009), pages 106–111, Boulder,…
  19. Saurav Kadavath, Tom Conerly, Amanda Askell, Tom Henighan, Dawn Drain, Ethan Perez, Nicholas Schiefer, Zac Hatfield-Dodds, Nova DasSarma, Eli Tran-Johnson, Scott Johnston, Sheer El-Showk, Andy Jones, Nelson Elhage, Tristan Hume, Anna Chen, Yuntao Bai, Sam Bowman, Stanislav Fort, and 17 others. 2022.… find
  20. Christopher Schläufer, Jan Künzler, Michael D. Jones, and 1 others. 2022. The narrative policy framework: A traveler’s guide to policy stories. Politische Viertel-jahresschrift, 63:249–273. find
  21. Alina Landowska, Katarzyna Budzynska, and He Zhang. 2024. Quantitative and qualitative analysis of moral foundations in argumentation - argumentation. find
  22. Nailia Mirzakhmedova, Johannes Kiesel, Milad Alshomary, Maximilian Heinrich, Nicolas Handke, Xiaoni Cai, Valentin Barriere, Doratossadat Dastgheib, Omid Ghahroodi, MohammadAli SadraeiJavaheri, Ehsaneddin Asgari, Lea Kawaletz, Henning Wachsmuth, and Benno Stein. 2024. The touché23-ValueEval dataset f… find
  23. Ines Rehbein, Ines Reinig, and Simone Paolo Ponzetto. 2025. Moral framing in politics (MFiP): A new resource and models for moral framing. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 34631–34651, Suzhou, China. Association for Computational Lingui… find
  24. Stefan Schweter and Alan Akbik. 2020. Flert: Document-level features for named entity recognition. Preprint, arXiv:2011.06993. — [arXiv:2011.06993](https://arxiv.org/abs/2011.06993) find
  25. Isabel Segura-Bedmar, Paloma Martínez, and María Herrero-Zazo. 2013. SemEval-2013 task 9 : Extraction of drug-drug interactions from biomedical texts (DDIExtraction 2013). In Second Joint Conference on Lexical and Computational Semantics (*SEM), Volume 2: Proceedings of the Seventh International Wor…
  26. Wei Li Seow, Iti Chaturvedi, Andrew Hogarth, and 1 others. 2025. A review of named entity recognition: from learning methods to modelling paradigms and tasks. Artificial Intelligence Review, 58:315. find
  27. Ottokar Tilk, Vera Demberg, Asad Sayeed, Dietrich Klakow, and Stefan Thater. 2016. Event participant modelling with neural networks. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 171–182, Austin, Texas. Association for Computational Linguistics. find
  28. A.2 Language model identifiers Zeliang Tong, Zhuojun Ding, and Wei Wei. 2025. Evo-Prompt: Evolving prompts for enhanced zero-shot named entity recognition with large language models. In Proceedings of the 31st International Conference on Computational Linguistics, pages 5136– 5153, Abu Dhabi, UAE. A… find
  29. Table 6 shows the short names used in this paper and their corresponding full model version identifiers. find
  30. Jackson Trager, Alireza S. Ziabari, Aida Mostafazadeh Davani, Preni Golazazian, Farzan Karimi-Malekabadi, Ali Omrani, Zhihe Li, Brendan Kennedy, Nils Karl Reimer, Melissa Lopez Reyes, Kelsey Cheng, Mellow Wei, Christina Merrifield, Arta Khosravi, Eva Luis Álvarez, and Morteza Dehghani. 2022. The mor…
  31. Short Name gpt-5-mini claude-3-5-haiku A.3 Selection of best models for our pipeline To identify the best-performing models for inclusion in our pipeline, we use the results from Section 4, Section 5.1, and the following formula: Pipeline Rec = Recall PD · Recall PC Zikang Zhang, Wangjie You, Tianci… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sommer-2026-who-plays-which-role-protagonist/