Elamur Research⌕
☰
home / papers / 2026 / Baselines for Detection and Classification of Discourse Presentation in English Narrative

Baselines for Detection and Classification of Discourse Presentation in English Narrative

Paper · 2026 · Reinaldo Di Polo, Mustafa Ocal and Mark Finlayson

Summary

Discourse presentation is the way a text conveys the speech, writing or thought of someone other than the narrator. It matters for tasks that depend on correct attribution, such as quotation extraction, opinion mining and narrative analysis. Verbatim (direct) forms usually carry clear signals like quotation marks. Indirect forms often lack such signals and blur the narrator's and the character's voices. The authors present what they describe as the first supervised evaluation covering the full range of discourse-presentation types in Semino & Short's SW&TP corpus.

First, they cleaned the corpus. They removed duplicated texts and repaired missing or inconsistent labels using a semi-automatic procedure followed by manual checking. They then reduced the original 25-tag scheme to a five-way clause classification task. On this task they compared four approaches: a CNN, a prompted Claude Sonnet 4.6, BERT without fine-tuning, and fine-tuned BERT. Fine-tuned BERT performed best on every class, but indirect categories remained the weak spot for all systems.

The cleaned data and code are publicly released, which gives future work a reproducible baseline.

Key points

  • Data: the SW&TP corpus contains 120 texts of roughly 2,000 words each, totalling 258,348 words and 14,380 sentences of late-20th-century British English. It is balanced across prose fiction, news reports and (auto)biography.
  • Labels: the 25 SW&TP tags are collapsed into five classes: speech and writing are grouped together and split by form into a direct and an indirect class, thought is likewise divided into Direct Thought and Indirect Thought, and every remaining tag falls under Narrative.
  • Overall F1 scores:
    • CNN: 0.43, with heavy confusion with Narrative.
    • Claude Sonnet 4.6: 0.71, and it fails entirely on Indirect Thought.
    • BERT without fine-tuning: 0.81.
    • Fine-tuned BERT with class-weighted loss: 0.88.
  • Hardest class: Indirect Speech & Writing is the most difficult category even for the best model (F1 = 0.60). It is frequently mistaken for Narrative.
  • Release and next steps: the corrected dataset and code are available at DOI 10.34703/gzx1-9v95/B8NPYV. Proposed future work includes:
    • using cross-sentence context;
    • span-based extraction of source, cue and content;
    • scaling to the finer-grained tagset;
    • per-genre analysis.

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

🎓Florida International UniversityMiami · United States8 works · 777 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 2Research subjects and languages the work is about.English · 2,263 worksDiscourse analysis · 466 works

Cite

Cite as Reinaldo Di Polo, Mustafa Ocal and Mark Finlayson. 2026. Baselines for Detection and Classification of Discourse Presentation in English Narrative. Proceedings of the 2nd Joint Workshop on Computational Approaches to Discourse, Context and Document-Level Inferences and Computational Models of Reference, Anaphora and Coreference (CODI-CRAC 2026).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.codi-1.3

@inproceedings{polo2026baselines,
  title = {{Baselines for Detection and Classification of Discourse Presentation in English Narrative}},
  author = {Reinaldo {Di Polo}, Mustafa Ocal and Mark Finlayson},
  booktitle = {Proceedings of the 2nd Joint Workshop on Computational Approaches to Discourse, Context and Document-Level Inferences and Computational Models of Reference, Anaphora and Coreference (CODI-CRAC 2026)},
  year = {2026},
  doi = {10.18653/v1/2026.codi-1.3},
  url = {https://doi.org/10.18653/v1/2026.codi-1.3}
}

References

24 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Geoffrey N Leech and Mick Short. 2007. Style in fiction: A linguistic introduction to English fictional prose. 13. Pearson Education. find
  2. Annelen Brunner. 2013. Automatic recognition of speech, thought, and writing representation in german narrative texts. Literary and linguistic computing, 28(4):563–575. ↗
  3. Brian McHale. 1978. Free indirect discourse: A survey of recent accounts. Poetics and Theory of Literature, 3:249–287. find
  4. Annelen Brunner, Stefan Engelberg, Fotis Jannidis, Ngoc Duyen Tanja Tu, and Lukas Weimer. 2020a. Corpus redewiedergabe. In Proceedings of the Twelfth Language Resources and Evaluation Conference, pages 803–812. find
  5. Grace Muzny, Michael Fang, Angel Chang, and Dan Jurafsky. 2017. A two-stage sieve approach for quote attribution. In Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers, pages 460–470. find
  6. Annelen Brunner, Ngoc Duyen Tanja Tu, Lukas Weimer, and Fotis Jannidis. 2020b. To bert or not to bert-comparing contextual embeddings in a deep learning architecture for the automatic recognition of four types of speech, thought and writing representation. In SwissText/KONVENS. find
  7. Neal R Norrick. 2016. Indirect reports, quotation and narrative. In Indirect reports and pragmatics: Interdisciplinary studies, pages 93–113. Springer. ↗
  8. Sean Papay and Sebastian Padó. 2019. Quotation detection and classification with a corpus-agnostic model. In Proceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2019), pages 888–894, Varna, Bulgaria. IN-COMA Ltd. find
  9. Carmen Rosa Caldas-Coulthard. 1994. On reporting reporting: The representation of speech in factual and factional narratives. In Malcolm Coulthard, editor, Advances in Written Text Analysis, pages 295–308. Routledge, London. find
  10. Florian Coulmas. 1986. Reported speech: Some general issues. Direct and indirect speech, 31:1–28. find
  11. Sean Papay and Sebastian Padó. 2020. Riqua: A corpus of rich quotation annotation for english literary text. In Proceedings of the Twelfth Language Resources and Evaluation Conference, pages 835–841.
  12. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805. ↗
  13. Silvia Pareti, Tim O’keefe, Ioannis Konstas, James R Curran, and Irena Koprinska. 2013. Automatically detecting and attributing indirect quotations. In Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 989–999. find
  14. David Elson, Nicholas Dames, and Kathleen McKeown. 2010. Extracting social networks from literary fiction. In Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, pages 138–147, Uppsala, Sweden. Association for Computational Linguistics. ↗
  15. Elena Semino and Mick Short. 2004. Corpus stylistics: Speech, writing and thought presentation in a corpus of English writing. Routledge. find
  16. Monika Fludernik. 1993. The Fictions of Language and the Languages of Fiction. Routledge, London. find
  17. Elena Semino, Mick Short, and Jonathan Culpeper. 1997. Using a corpus to test a model of speech and thought presentation. Poetics, 25(1):17–43. ↗
  18. Kunihiko Fukushima. 1980. Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position. Biological Cybernetics. ↗
  19. Masaki Shibata. 2021. Reported speech as persuasion: A discourse analysis of japanese journalism. Japanese Studies, 41(2):221–239. ↗
  20. Matt Gardner, Joel Grus, Mark Neumann, Oyvind Tafjord, Pradeep Dasigi, Nelson F Liu, Matthew E Peters, Michael Schmitz, and Luke Zettlemoyer. 2018. Allennlp: A deep semantic natural language processing platform. In Proceedings of workshop for NLP open source software (NLP-OSS), pages 1–6.
  21. Linda R. Waugh. 1995. Reported speech in journalistic discourse: The relation of function and text. Text, 15(1):129–173. ↗
  22. Yonghui Wu, Mike Schuster, Zhifeng Chen, Quoc V. Le, Mohammad Norouzi, Wolfgang Macherey, Maxim Krikun, Yuan Cao, Qin Gao, Klaus Macherey, Jeff Klingner, Apurva Shah, Melvin Johnson, Xiaobing Liu, Lukasz Kaiser, Stephan Gouws, Yoshikiyo Kato, Taku Kudo, and Hideto Kazawa. 2016. Google’s neural machine translation system: Bridging the gap between human and machine translation. arXiv preprint arXiv:1609.08144. ↗
  23. Ralf Krestel, Sabine Bergler, and René Witte. 2008. Minding the source: Automatic tagging of reported speech in newspaper articles. In Proceedings of the Sixth International Conference on Language Resources and Evaluation (LREC’08), Marrakech, Morocco. European Language Resources Association (ELRA). find
  24. Geoffrey N. Leech and Mick Short. 1981. Style in Fiction: A Linguistic Introduction to English Fictional Prose. Longman, London. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/polo-2026-baselines-detection-classification-discourse-presentation/