Elamur Research⌕
☰
home / papers / 2026 / Blaming Across the Aisle: Political Contrasting and Blame Attribution in the Danish Parliament

Blaming Across the Aisle: Political Contrasting and Blame Attribution in the Danish Parliament

Paper · 2026 · Markus Lundsfryd Jensen, Rune Egeskov Trust, Kenneth Christian Enevoldsen et al.

Summary

Commentators often claim that political talk is getting harsher, but evidence for this is thin, and most of it comes from social media rather than parliaments. The authors study blame in Danish parliamentary speeches from 1997 to 2026. They define blame as a statement that assigns cause and carries negative sentiment. They merged ParlSpeechV2 with newer transcripts, split the speeches into sentences with DaCy, and machine-translated a sample into English. An English zero-shot NLI model, DEBATE, then labelled the translated sample using several prompt templates. Those silver labels were used to fine-tune mmBERT into BlameBERT, which reaches F1 0.80 on a manually checked test set. Multilevel count models applied to the classifier's outputs show a U-like ("banana") curve over time: blame falls until about 2016 and rises clearly from 2019 onward. Opposition parties blame far more than governing parties, which the authors call "political contrasting". This gap depends on ideology, with right-wing and more extreme parties blaming more. The study also offers a reusable, low-annotation recipe for languages with fewer resources.

Key points

  • Corpus: 6,553,133 sentences, of which 5,598,994 remained after filtering. A random 500,500 were used for development and the rest for inference.
  • Silver labelling: Opus-MT-da-en translated the development sentences. DEBATE scored each one under five hypothesis templates with the labels blame, praise and neutral. A sentence counted as blame when that label had probability ≥ .80 and outscored the other two labels.
  • DIAL datasets: "Datasets of Increasing Agreement Levels" (DIAL-1 to DIAL-5) mark a sentence positive when at least n templates agree. Blame prevalence ranges from 1.71% (DIAL-1) down to 0.50% (DIAL-5).
  • Gold test set: 500 upsampled sentences (250 blame, 250 non-blame) were annotated by two authors. Agreement was 84.8% (Cohen's κ = .676), and the 424 items both agreed on were kept, 148 of them blame.
  • Training: BlameBERT is mmBERT with LoRA (rank 64, alpha 128) and focal loss, selected by validation MCC. The released model and dataset are on Hugging Face. The main findings hold across classification thresholds, except one interaction, which loses significance at the strictest threshold.

Interesting ideas

  • I1. Agreement among several prompt templates serves as a tunable strictness setting: one silver set is produced at each consensus level, a model is trained on each, and a small gold set decides which level works best. This avoids guessing the right noise/coverage trade-off up front.
  • I2. Blame is treated as a composite (a causal attribution plus negative evaluation) rather than as plain negative sentiment. This makes the label clearer and less biased by assumptions about the speaker.
  • I3. The zero-shot label set pairs an antonym ("praise") with a "neutral" escape option. The model is therefore never forced into a binary choice, and mixed or unclear sentences are absorbed by neutral.
  • I4. The gold test set is built by drawing balanced samples from the silver positives and negatives and keeping only items both annotators agreed on. This gives a cheap, class-balanced evaluation set even for a rare phenomenon.
  • I5. The authors admit that their classifier finds blame but not its target. A party blaming the EU or a pandemic is a different phenomenon from one attacking domestic rivals, so a label without a target can mislead.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (learn): Silver trees and frames could be grouped into tiers by how many witness parsers or LLM-teacher prompts agree, and gold-set results would choose which tier to train on.
  • I2 (frames): Blame could be detected structurally, as a causation frame whose agent receives a negative evaluation, which keeps each detection explainable instead of relying on a sentiment score.
  • I3 (frames): Speech-act and irony classifiers in prag could include an explicit contrastive act plus a neutral/mixed outcome, so ambiguous utterances are not pushed into a label.
  • I4 (learn): Rare constructions, such as blame or concessive frames, could get small balanced test sets drawn from silver output and checked by two annotators, used as negative controls on the silver pipeline.
  • I5 (world): Each blame frame should be linked to a resolved target entity in the world of the text, so questions like "who blames whom" can be answered and traced back to the sentence.

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

 Fields, topics and concepts

📚 Shelves 6The 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 worksSocial Sciences · 1,772 worksGeneral Social Sciences · 227 works

Cite

Cite as Markus Lundsfryd Jensen, Rune Egeskov Trust, Kenneth Christian Enevoldsen and Sara Kolding. 2026. Blaming Across the Aisle: Political Contrasting and Blame Attribution in the Danish Parliament. arXiv:2609.26346.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2609.26346 · arXiv 2609.26346

@misc{jensen2026blaming,
  title = {{Blaming Across the Aisle: Political Contrasting and Blame Attribution in the Danish Parliament}},
  author = {Markus Lundsfryd Jensen, Rune Egeskov Trust, Kenneth Christian Enevoldsen and Sara Kolding},
  year = {2026},
  doi = {10.48550/arxiv.2609.26346},
  eprint = {2609.26346},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2609.26346}
}

References

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

  1. Nathalie Damgaard Frisch. 2022. Hvis politikerne ønsker en bedre tone i debatten, skal de måske starte med sig selv, viser forskningen. find
  2. Anju, Inderdeep Kaur Aulakh, and Raj Kumari. 2024. The impact of employing sampling-based strategies to address the issue of an imbalanced dataset in hate speech and offensive language categorization. In 2024 15th International Conference on Computing Communication and Networking Technologies (ICC-CNT), pages 1–7. find
  3. S Frisch and S Kelly. 2013. Politics to the extreme: American political institutions in the twenty-first century. Springer. find
  4. R Kelly Garrett, Shira Dvir Gvirsman, Benjamin K Johnson, Yariv Tsfati, Rachel Neo, and Aysenur Dal. 2014. Implications of pro-and counterattitudinal information exposure for affective polarization. Human communication research, 40(3):309–332. ↗
  5. Camilla Alexandra Barton. 2023. Er tonen i Folketinget for hård? Det er der delte meninger om - TV 2. nyheder.tv2.dk. find
  6. Emma Joisten Becker. 2021. "Bagsiden af medaljen ved sociale medier er ved at være større end forsiden". find
  7. Søren Peter Hansen. 2024. Tænketanken Prospekt: Vi har brug for at genfinde etikken i politik - Altinget. Section: Etik og Tro. find
  8. Irina Bigoulaeva, Viktor Hangya, Iryna Gurevych, and Alexander Fraser. 2022. Addressing the challenges of cross-lingual hate speech detection. arXiv preprint arXiv:2201.05922. ↗
  9. Tim Heinkelmann-Wild and Bernhard Zangl. 2020. Multilevel blame games: Blame-shifting in the european union. Governance, 33(4):953–969. ↗
  10. Francesco Bilotta, Alberto Binetti, and Giacomo Manferdini. 2025. Blameocracy: Causal Attribution in Political Communication. arXiv preprint. ArXiv:2504.06550 [econ]. ↗
  11. Markus Hinterleitner. 2020. Blame Games in the Political Sphere, page 18–39. Cambridge Studies in Comparative Public Policy. Cambridge University Press. find
  12. Maiken Brusgaard. 2022. Er tonen i dansk politik for hård? Her er otte eksempler på, at den altid har været det - TV 2. nyheder.tv2.dk. find
  13. Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2021. LoRA: Low-Rank Adaptation of Large Language Models. arXiv preprint. ArXiv:2106.09685. ↗
  14. Michael Burnham, Kayla Kahn, Ryan Yang Wang, and Rachel X. Peng. 2026. Political debate: Efficient zero-shot and few-shot classifiers for political text. Political Analysis, 34(3):329–343. ↗
  15. Kate Kenski and Kathleen Hall Jamieson. 2017. The Oxford handbook of political communication. Oxford University Press. find
  16. Tobias Burst, Pola Lehmann, Simon Franzmann, Denise Al-Gaddooa, Christoph Ivanusch, Sven Regel, Felicia Riethmüller, Bernhard Weßels, Lisa Zehnter, Wissenschaftszentrum Berlin Für Sozialforschung (WZB), and Institut Für Demokratieforschung Göttingen (IfDem). 2024. Manifestoberta. find
  17. Moritz Laurer, Wouter van Atteveldt, Andreu Casas, and Kasper Welbers. 2023. Building Efficient Universal Classifiers with Natural Language Inference. arXiv preprint. ↗
  18. Troels Bøggild and Carsten Jensen. 2025. When politicians behave badly: Political, democratic, and social consequences of political incivility. American Journal of Political Science, 69(3):1064–1081. ↗
  19. Suvi Lehtosalo and John Nerbonne. 2024. Detecting emotional polarity in Finnish parliamentary proceedings. In Proceedings of the 4th Workshop on Computational Linguistics for the Political and Social Sciences: Long and short papers, pages 90–100, Vienna, Austria. Association for Computational Linguistics. find
  20. Paul Chilton. 2004. Analysing political discourse: Theory and practice. routledge. find
  21. Shuailong Liang, Olivia Nicol, and Yue Zhang. 2019. Who Blames Whom in a Crisis? Detecting Blame Ties from News Articles Using Neural Networks. Proceedings of the AAAI Conference on Artificial Intelligence, 33(01):655–662. find
  22. Christian Elmelund-Præstekær. 2010. Beyond american negativity: Toward a general understanding of the determinants of negative campaigning. European Political Science Review, 2(1):137–156. ↗
  23. Bertram Malle, Steve Guglielmo, and Andrew Monroe. 2014. A Theory of Blame. Psychological Inquiry, 25:147–186. ↗
  24. Kenneth Enevoldsen, Lasse Hansen, and Kristoffer Nielbo. 2021. DaCy: A Unified Framework for Danish NLP. arXiv preprint. ArXiv:2107.05295. ↗
  25. Bertram F. Malle, Steve Guglielmo, and Andrew E. Monroe. 2013. Moral, cognitive, and social: The nature of blame. In Social thinking and interpersonal behavior, Sydney symposium of social psychology, pages 313–331. Psychology Press, New York, NY, US. find
  26. Jeremy A Frimer, Harinder Aujla, Matthew Feinberg, Linda J Skitka, Karl Aquino, Johannes C Eichstaedt, and Robb Willer. 2023. Incivility is rising among american politicians on twitter. Social Psychological and Personality Science, 14(2):259–269. ↗
  27. Marc Marone, Orion Weller, William Fleshman, Eugene Yang, Dawn Lawrie, and Benjamin Van Durme. 2025. mmbert: A modern multilingual encoder with annealed language learning. Preprint, arXiv:2509.06888. ↗
  28. Sven-Oliver Proksch and Jonathan B Slapin. 2014. The politics of parliamentary debate: Parties, rebels and representation. Cambridge University Press. find
  29. Qwen Team. 2026. Qwen3.5: Towards native multimodal agents. find
  30. Lilliana Mason. 2015. “i disrespectfully agree”: The differential effects of partisan sorting on social and issue polarization. American journal of political science, 59(1):128–145. find
  31. R Core Team. 2025. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. find
  32. Maeve McGillycuddy, David I. Warton, Gordana Popovic, and Benjamin M. Bolker. 2025. Parsimoniously fitting large multivariate random effects in glmmTMB. Journal of Statistical Software, 112(1):1– 19. ↗
  33. Heng Rathpisey and Teguh Bharata Adji. 2019. Handling imbalance issue in hate speech classification using sampling-based methods. In 2019 5th International Conference on Science in Information Technology (ICSITech), pages 193–198. find
  34. Michal Mochtak, Peter Rupnik, Taja Kuzman, and Nikola Ljubešić. 2025. Parlasent: mapping sentiment in political discourse with large language models. Political Research Exchange, 7(1):2508377. ↗
  35. Christian Rauh and Jan Schwalbach. 2020. Corp_folketing_v2.rds. In The ParlSpeech V2 data set: Full-text corpora of 6.3 million parliamentary speeches in the key legislative chambers of nine representative democracies. Harvard Dataverse. Version Number: V1. find
  36. Domenico Montanaro. 2018. Poll: Nearly 4 in 5 voters concerned incivility will lead to violence. NPR. find
  37. Ludovic Rheault, Kaspar Beelen, Christopher Cochrane, and Graeme Hirst. 2016. Measuring emotion in parliamentary debates with automated textual analysis. PloS one, 11(12):e0168843. ↗
  38. Ashley Muddiman. 2017. Personal and public levels of political incivility. International Journal of Communication, 11:21–21. find
  39. Costanza Navarretta and Dorte Haltrup Hansen. 2024. Government and opposition in Danish parliamentary debates. In Proceedings of the IV Workshop on Creating, Analysing, and Increasing Accessibility of Parliamentary Corpora (ParlaCLARIN) @ LREC-COLING 2024, pages 154–162, Torino, Italia. ELRA and ICCL. find
  40. Anna Ristilä, Otto Tarkka, Veronika Laippala, and Kimmo Elo. 2026. Hopes and Fears – Emotion Distribution in the Topic Landscape of Finnish Parliamentary Speech 2000-2020. Preprint, arXiv:2601.20424. ↗
  41. Jan Rovny, Jonathan Polk, Ryan Bakker, Liesbet Hooghe, Seth Jolly, Gary Marks, Marco Steenbergen, and Milada Anna Vachudova. 2025. The 2024 chapel hill expert survey on political party positioning in europe: Twenty-five years of party positional data. Electoral Studies, 97:102981. ↗
  42. Costanza Navarretta, Dorte Haltrup Hansen, and Bart Jongejan. 2022. Immigration in the manifestos and parliament speeches of Danish left and right wing parties between 2009 and 2020. In Proceedings of the Workshop ParlaCLARIN III within the 13th Language Resources and Evaluation Conference, pages 71–80, Marseille, France. European Language Resources Association. find
  43. Gijs Schumacher, Daniel Hansen, Mariken A.C.G. van der Velden, and Sander Kunst. 2019. A new dataset of Dutch and Danish party congress speeches. Research & Politics, 6(2):2053168019838352. ↗
  44. Kunwoo Park, Zhufeng Pan, and Jungseock Joo. 2021. Who blames or endorses whom? entity-to-entity directed sentiment extraction in news text. In Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 4091–4102, Online. Association for Computational Linguistics. find
  45. Weber Shandwick and Powell Tate. 2019. Civility in america 2019: Solutions for tomorrow. Weber Shandwick We Solve. find
  46. David Shoemaker. 2015. Responsibility from the Margins. Oxford University Press (UK). ↗
  47. Lukas Pätz, Moritz Beyer, Jannik Späth, Lasse Bohlen, Patrick Zschech, Mathias Kraus, and Julian Rosenberger. 2025. Analyzing German Parliamentary Speeches: A Machine Learning Approach for Topic and Sentiment Classification. Preprint, arXiv:2508.03181. ↗
  48. David Shoemaker and Manuel Vargas. 2021a. Moral torch fishing: A signaling theory of blame. Noûs, 55(3):581–602. find
  49. David Shoemaker and Manuel Vargas. 2021b. Moral torch fishing: A signaling theory of blame. Noûs, 55(3):581–602. find
  50. Douglas W Portmore. 2022. A comprehensive account of blame: Self-blame, non-moral blame, and blame for the non-voluntary. find
  51. Angela Smith. 2012. Moral blame and moral protest. ↗
  52. G Bingham Powell Jr. 2004. Political representation in comparative politics. Annu. Rev. Polit. Sci., 7(1):273– 296. ↗
  53. Hans-Henrik Busk Stie. 2021. Danskere siger fra og blokerer eller unfriender ’venner’ på Facebook - TV 2. nyheder.tv2.dk. Section: Stop hadbeskeder. find
  54. Wolf Susan, Jay Wallace, Kumar Rahul, and Freeman Samuel. 2011. Blame, Italian Style. Reasons and Recognition: Essays on the Philosophy of TM Scanlon, pages 332–47. ↗
  55. Yannis Theocharis, Pablo Barberá, Zoltán Fazekas, and Sebastian Adrian Popa. 2020. The dynamics of political incivility on twitter. Sage Open, 10(2):2158244020919447. ↗
  56. Jörg Tiedemann, Mikko Aulamo, Daria Bakshandaeva, Michele Boggia, Stig-Arne Grönroos, Tommi Nieminen, Alessandro Raganato, Yves Scherrer, Raul Vazquez, and Sami Virpioja. 2023. Democratizing Neural Machine Translation with OPUS-MT. arXiv preprint. ArXiv:2212.01936 [cs]. ↗
  57. Neal Tognazzini and D. Justin Coates. 2024. Blame. In Edward N. Zalta and Uri Nodelman, editors, The Stanford Encyclopedia of Philosophy, fall 2025 edition. Metaphysics Research Lab, Stanford University. find
  58. Jonathan Van’t Riet and Aart Van Stekelenburg. 2022. The effects of political incivility on political trust and political participation: A meta-analysis of experimental research. Human Communication Research, 48(2):203–229. ↗
  59. Jay Wallace. 2011. Dispassionate opprobrium: On blame and the reactive sentiments. find
  60. Shawn Tinghao Wang. 2024. Rethinking Functionalist Accounts of Blame. The Journal of Ethics, 28(4):607–623. Williams. Praise and Blame | Internet Encyclopedia of Philosophy. ↗
  61. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pier-ric Cistac, Tim Rault, Rémi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, and 3 others. 2020. HuggingFace’s Transformers: State-of-the-art Natural Language Processing. arXiv preprint. ArXiv:1910.03771 [cs]. ↗
  62. Lori Young and Stuart Soroka. 2012. Affective news: The automated coding of sentiment in political texts. Political communication, 29(2):205–231. ↗
  63. Yanzhao Zhang, Mingxin Li, Dingkun Long, Xin Zhang, Huan Lin, Baosong Yang, Pengjun Xie, An Yang, Dayiheng Liu, Junyang Lin, Fei Huang, and Jingren Zhou. 2025. Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models. arXiv preprint arXiv:2506.05176. ↗
  64. Jingming Zhuo, Songyang Zhang, Xinyu Fang, Haodong Duan, Dahua Lin, and Kai Chen. 2024. ProSA: Assessing and Understanding the Prompt Sensitivity of LLMs. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 1950–1976, Miami, Florida, USA. Association for Computational Linguistics. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/jensen-2026-blaming-across-aisle-political-contrasting/