Elamur Research⌕
☰
home / papers / 2026 / PuDy at FadeIT: Enhancing Fine-Grained Fallacy Detection with Hierarchy-Aware Training and Contextual Enrichment

PuDy at FadeIT: Enhancing Fine-Grained Fallacy Detection with Hierarchy-Aware Training and Contextual Enrichment

Paper · 2026 · Duy Dang Phu, Son Bui Hong and Dang Van Thin

Summary

The paper presents the top-ranked system in FadeIT's Subtask B, a shared task held at EVALITA 2026. The task is to locate and label fallacious text spans in Italian social media posts, where spans can overlap and labels follow a hierarchical taxonomy. Training data are scarce, and scoring uses Soft F1, which gives partial credit when a prediction matches the parent category of the gold label.

The authors treat the task as multi-label span classification on top of UmBERTo. Each candidate span is represented by its boundary token embeddings plus a width embedding, and the model is trained with Focal Loss.

They add two components:

  • Label propagation: whenever a child label is active, its parent label is raised to at least a minimum target value, which matches training to how the metric scores predictions.
  • Contextual Enrichment: the Gemini API rewrites only the non-fallacious context and leaves the annotated spans untouched.

The system ranked first with a Soft F1 of 50.92. The results suggest that metric-aware targets and augmentation that keeps the labelled spans intact help in low-resource argument mining.

Key points

  • Architecture: UmBERTo encoder; candidate spans are capped at 20 tokens, and each is represented as [h_i; h_j; width embedding] with a sigmoid multi-label classifier.
  • Augmentation comparison: a second, "span-relaxed" variant also paraphrases the fallacious spans. It suffered from semantic drift and label noise, so Contextual Enrichment gave steadier precision and the best Soft Micro F1.
  • Best run (Run 1): Soft P/R/F1 of 43.76/60.88/50.92, against a baseline Soft F1 of 6.25. Under strict evaluation, however, team TiGRO scored higher (42.13 F1 vs. PuDy's 31.97).
  • Failed categories: six fallacy types scored 0.00 F1 in all runs: Causal-oversimplification, Cherry-picking, Circular-reasoning, False-analogy, False-dilemma and Strawman. The authors attribute this to their usually long spans exceeding the 20-token limit.
  • Strongest categories: Slogan (up to 64.07 F1) and Name-calling-or-labelling (about 57 F1), both of which tend to be expressed in short spans.

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 2The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 works

Cite

Cite as Duy Dang Phu, Son Bui Hong and Dang Van Thin. 2026. PuDy at FadeIT: Enhancing Fine-Grained Fallacy Detection with Hierarchy-Aware Training and Contextual Enrichment. Proceedings of the Ninth Evaluation Campaign of Natural Language Processing and Speech Tools for Italian. Final Workshop (EVALITA 2026).

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{phu2026pudy,
  title = {{PuDy at FadeIT: Enhancing Fine-Grained Fallacy Detection with Hierarchy-Aware Training and Contextual Enrichment}},
  author = {Duy Dang Phu, Son Bui Hong and Dang {Van Thin}},
  booktitle = {Proceedings of the Ninth Evaluation Campaign of Natural Language Processing and Speech Tools for Italian. Final Workshop (EVALITA 2026)},
  year = {2026},
  url = {https://aclanthology.org/2026.evalita-1.65/}
}

References

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

  1. A. Ramponi, S. Tonelli, FadeIT at EVALITA 2026: Overview of the fallacy detection in italian social media texts task, in: Proceedings of the Ninth Evaluation Campaign of Natural Language Processing and Speech Tools for Italian. Final Workshop (EVALITA 2026), CEUR.org, Bari, Italy, 2026.
  2. F. Cutugno, A. Miaschi, A. P. Aprosio, G. Rambelli, L. Siciliani, M. A. Stranisci, Evalita 2026: Overview of the 9th evaluation campaign of natural language processing and speech tools for italian, in: Proceedings of the Ninth Evaluation Campaign of Natural Language Processing and Speech Tools for I… find
  3. Z. Jin, A. Lalwani, T. Vaidhya, X. Shen, Y. Ding, Z. Lyu, M. Sachan, R. Mihalcea, B. Schölkopf, Logical fallacy detection, in: Y. Goldberg, Z. Kozareva, Y. Zhang (Eds.), Findings of the Association for Computational Linguistics: EMNLP 2022, Association for Computational Linguistics, Abu Dhabi, Unite…
  4. C. Helwe, T. Calamai, P.-H. Paris, C. Clavel, F. Suchanek, MAFALDA: A benchmark and comprehensive study of fallacy detection and classification, in: K. Duh, H. Gomez, S. Bethard (Eds.), Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics:… find
  5. A. Ramponi, A. Daffara, S. Tonelli, Fine-grained fallacy detection with human label variation, in: L. Chiruzzo, A. Ritter, L. Wang (Eds.), Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1… find
  6. E. Cantín Larumbe, A. Chust Vendrell, Argumentative fallacy detection in political debates, in: E. Chistova, P. Cimiano, S. Haddadan, G. Lapesa, R. Ruiz-Dolz (Eds.), Proceedings of the 12th Argument mining Workshop, Association for Computational Linguistics, Vienna, Austria, 2025, pp. 369–373. URL:… find
  7. Y. Lei, R. Huang, Boosting logical fallacy reasoning in LLMs via logical structure tree, in: Y. Al-Onaizan, M. Bansal, Y.-N. Chen (Eds.), Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, Association for Computational Linguistics, Miami, Florida, USA, 2024, pp.… find
  8. M. Joshi, D. Chen, Y. Liu, D. S. Weld, L. Zettlemoyer, O. Levy, SpanBERT: Improving pre-training by representing and predicting spans, Transactions of the Association for Computational Linguistics 8 (2020) 64–77. URL: https://aclanthology.org/2020.tacl-1.5/. doi: 10.1162/tacl_a_00300 . ↗
  9. J. Fu, X. Huang, P. Liu, SpanNER: Named entity re-/recognition as span prediction, in: C. Zong, F. Xia, W. Li, R. Navigli (Eds.), Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume… find
  10. D. Wadden, U. Wennberg, Y. Luan, H. Hajishirzi, Entity, relation, and event extraction with contextualized span representations, in: K. Inui, J. Jiang, V. Ng, X. Wan (Eds.), Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Confere…
  11. S. Kobayashi, Contextual augmentation: Data augmentation by words with paradigmatic relations, in: M. Walker, H. Ji, A. Stent (Eds.), Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers…
  12. H. Dai, Z. Liu, W. Liao, X. Huang, Y. Cao, Z. Wu, L. Zhao, S. Xu, F. Zeng, W. Liu, N. Liu, S. Li, D. Zhu, H. Cai, L. Sun, Q. Li, D. Shen, T. Liu, X. Li, AugGPT: Leveraging ChatGPT for Text Data Augmentation , IEEE Transactions on Big Data 11 (2025) 907–918. URL: https://doi.ieeecomputersociety. org/… — [doi:10.1109/tbdata.2025.3536934](https://doi.org/10.1109/tbdata.2025.3536934) ↗
  13. D. Kaushik, E. Hovy, Z. C. Lipton, Learning the difference that makes a difference with counterfactually augmented data, International Conference on Learning Representations (ICLR) (2020). find
  14. L. Parisi, S. Francia, P. Magnani, Umberto: an italian language model trained with whole word masking, https://github.com/musixmatchresearch/umberto, 2020. ↗
  15. T.-Y. Lin, P. Goyal, R. Girshick, K. He, P. Dollár, Focal loss for dense object detection, in: 2017 IEEE International Conference on Computer Vision (ICCV), 2017, pp. 2999–3007. doi: 10.1109/ICCV. 2017.324 .
  16. J. Cegin, B. Pecher, J. Simko, I. Srba, M. Bielikova, P. Brusilovsky, Use random selection for now: Investigation of few-shot selection strategies in LLM-based text augmentation, in: C. Christodoulopoulos, T. Chakraborty, C. Rose, V. Peng (Eds.), Findings of the Association for Computational Linguis… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/phu-2026-pudy-fadeit-enhancing-fine-grained/