Elamur Research⌕
☰
home / papers / 2026 / NCL&HKU-NarrSim at SemEval-2026 Task 4: Aspect-Based Agents and Supervised Contrastive Embeddings for Narrative Similarity

NCL&HKU-NarrSim at SemEval-2026 Task 4: Aspect-Based Agents and Supervised Contrastive Embeddings for Narrative Similarity

Paper · 2026 · Jianfei Xu, Ting Zhu, Mingyang Chen et al.

Summary

This paper presents the NCL&HKU-NarrSim system for SemEval-2026 Task 4. The task asks which of two candidate stories is narratively closer to an anchor story, judged on deeper structure rather than shared wording. The authors propose one approach per track.

For Track A, ABNS-Agents is a two-stage LLM pipeline. A first agent converts every story into a fixed schema organized around the three narrative aspects the task defines, namely the outcomes, the abstract theme and the course of action. A second LLM then compares these aspects and picks the closer candidate.

For Track B, NSConE builds on supervised SimCSE with RoBERTa-large. It is fine-tuned on anchor/positive/negative triples with an InfoNCE-style loss, plus an auxiliary MLM term to limit catastrophic forgetting.

The official synthetic training set showed strong lexical overlap, which allowed shortcut solutions. The team therefore discarded it and trained only on the official development split.

ABNS-Agents reached 70.25% test accuracy and NSConE 68.5%. These results suggest that both structured reasoning and task-specific contrastive learning can capture narrative similarity beyond surface semantics.

Key points

  • The synthetic training data could be solved with over 98% accuracy using word overlap alone. All training and validation therefore used the official dev set, split 80/20.
  • In ABNS-Agents, the theme and the outcome are each summarized in one sentence, and the course of action as exactly three short events. The best setup used GPT-4o-mini for extraction and GPT-4o for the decision: 68.5% on dev and 70.25% on test.
  • A weighted decision strategy (66.0% dev) and GPT-5.2 for both stages (67.0% dev) did not beat the best configuration.
  • Over 20 encoders were benchmarked, including BERT, ModernBERT, DeBERTa-v3, Longformer, LaBSE, Sentence-T5 and StoryEmb. Sup-SimCSE-RoBERTa-large was chosen as the backbone because it was stable across settings; LaBSE was also strong but more sensitive to data scale.
  • NSConE raised validation accuracy from 67.5% to 72.5% over its baseline, reached 68.5% on test, and kept an STS-AVG of 82.29. The code is published on GitHub (jianfeixu95/NCL-SemEval2026-NarrSim).

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

🎓University of Hong KongPok Fu Lam · China89 works · 11,774 citations🎓Newcastle UniversityNewcastle upon Tyne · United Kingdom18 works · 619 citations

 Fields, topics and concepts

📚 Shelves 5The 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 worksComputer Vision and Pattern Recognition · 2,869 works

Cite

Cite as Jianfei Xu, Ting Zhu, Mingyang Chen and Huizhi(elly) Liang. 2026. NCL&HKU-NarrSim at SemEval-2026 Task 4: Aspect-Based Agents and Supervised Contrastive Embeddings for Narrative Similarity. Proceedings of the 20th International Workshop on Semantic Evaluation (2026).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.semeval-1.188

@inproceedings{xu2026ncl,
  title = {{NCL\&HKU-NarrSim at SemEval-2026 Task 4: Aspect-Based Agents and Supervised Contrastive Embeddings for Narrative Similarity}},
  author = {Jianfei Xu, Ting Zhu, Mingyang Chen and Huizhi(elly) Liang},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.188},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.188}
}

References

19 references, 12 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Eneko Agirre, Daniel Cer, Mona Diab, and Aitor Gonzalez-Agirre. 2012. SemEval-2012 task 6: A pilot on semantic textual similarity. In *SEM 2012: The First Joint Conference on Lexical and Computational Semantics – Volume 1: Proceedings of the main conference and the shared task, and Volume 2: Proceed…
  2. Prithviraj Ammanabrolu, Ethan Tien, Wesley Cheung, Zhaochen Luo, William Ma, Lara J. Martin, and Mark O. Riedl. 2020. Story Realization: Expanding Plot Events into Sentences. Proceedings of the AAAI Conference on Artificial Intelligence, 34(05):7375– 7382.
  3. Laura Banarescu, Claire Bonial, Shu Cai, Madalina Georgescu, Kira Griffitt, Ulf Hermjakob, Kevin Knight, Philipp Koehn, Martha Palmer, and Nathan Schneider. 2013. Abstract Meaning Representation for Sembanking. In Proceedings of the 7th Linguistic Annotation Workshop and Interoperability with Discou…
  4. Iz Beltagy, Matthew E. Peters, and Arman Cohan. 2020. Longformer: The Long-Document Transformer.
  5. Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter,… — Language Models are Few-Shot Learners
  6. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,…
  7. Tianyu Gao, Xingcheng Yao, and Danqi Chen. 2021. SimCSE: Simple contrastive learning of sentence embeddings. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 6894–6910, Online and Punta Cana, Dominican Republic. Association for Computational Linguistic…
  8. Hans Ole Hatzel, Ekaterina Artemova, Haimo Stiemer, Evelyn Gius, and Chris Biemann. 2026. SemEval-2026 Task 4: Narrative similarity and narrative representation learning. In Proceedings of the 20th International Workshop on Semantic Evaluation (SemEval-2026), San Diego, CA, USA. Association for Comp… find
  9. Prannay Khosla, Piotr Teterwak, Chen Wang, Aaron Sarna, Yonglong Tian, Phillip Isola, Aaron Maschinot, Ce Liu, and Dilip Krishnan. 2020. Supervised contrastive learning. In Proceedings of the 34th International Conference on Neural Information Processing Systems, NIPS ’20, Red Hook, NY, USA. Curran… find
  10. Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, and Amr Ahmed. 2020. Big bird: transformers for longer sequences. In Proceedings of the 34th International Conference on Neural Information Processing Systems, NIPS ’20, Red Hook, NY, USA. Curran Associates Inc.
  11. Liu Zhuang, Lin Wayne, Shi Ya, and Zhao Jun. 2021. A robustly optimized BERT pre-training approach with post-training. In Proceedings of the 20th Chinese National Conference on Computational Linguistics, pages 1218–1227, Huhhot, China. Chinese Information Processing Society of China.
  12. Ryan Kiros, Yukun Zhu, Ruslan Salakhutdinov, Richard S. Zemel, Antonio Torralba, Raquel Urtasun, and Sanja Fidler. 2015. Skip-thought vectors. In Proceedings of the 29th International Conference on Neural Information Processing Systems - Volume 2, NIPS’15, page 3294–3302, Cambridge, MA, USA. MIT Pre… find
  13. Lane Lawley, Gene Louis Kim, and Lenhart Schubert. 2019. Towards natural language story understanding with rich logical schemas. In Proceedings of the Sixth Workshop on Natural Language and Computer Science, pages 11–22, Gothenburg, Sweden. Association for Computational Linguistics. find
  14. Pengfei Liu, Weizhe Yuan, Jinlan Fu, Zhengbao Jiang, Hiroaki Hayashi, and Graham Neubig. 2023. Pre-train, prompt, and predict: A systematic survey of prompting methods in natural language processing. ACM Comput. Surv., 55(9).
  15. Matthew E. Peters, Mark Neumann, Mohit Iyyer, Matt Gardner, Christopher Clark, Kenton Lee, and Luke Zettlemoyer. 2018. Deep contextualized word representations. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Techno…
  16. Nils Reimers and Iryna Gurevych. 2019. Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 3982–3992,…
  17. Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Thomas L. Griffiths, Yuan Cao, and Karthik Narasimhan. 2023. Tree of thoughts: deliberate problem solving with large language models. In Proceedings of the 37th International Conference on Neural Information Processing Systems, NIPS ’23, Red Hook, NY… find
  18. Manzil Zaheer, Guru Guruganesh, Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Type Split Num Close to A Close to B Official Sample Dev Train 39 200 1900(LLM-generated) 24 99 957 15 101 943 find
  19. Ours Train Validation 160 40 85 16 75 24 *Table 3: Dataset statistics. The ours split is derived by partitioning the official dev split into training(160 instances) and validation(40 instances) sets.* Field Split Mean Median Std P10 P90 Min Max anchor_text anchor_text anchor_text text_a text_a text_… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/xu-2026-ncl-hku-narrsim-semeval-2026/