Elamur Research⌕
☰
home / papers / 2026 / HCMUS RepeatedGames at SemEval-2026 Task 12: CausalRAG: Synergizing Causal Graph Retrieval and Extended LoRA for Abductive Reasoning

HCMUS RepeatedGames at SemEval-2026 Task 12: CausalRAG: Synergizing Causal Graph Retrieval and Extended LoRA for Abductive Reasoning

Paper · 2026 · Duy Minh Dao Sy, Nguyen Tran, Trung Kiet Huynh et al.

Summary

SemEval-2026 Task 12 (Abductive Event Reasoning, AER) gives systems a real-world event, about ten retrieved English documents per topic, and four candidate explanations. Systems must choose the one or more options that best explain the event, and one option always says that none of the others applies. The HCMUS_RepeatedGames team built CausalRAG, which treats the task as multi-label classification instead of text generation.

Retrieval combines BM25 and Sentence-BERT dense scores. It also gives extra weight to chunks where regex patterns detect explicit cause–effect wording. For reasoning, the team fine-tunes Qwen2.5-32B-Instruct with 4-bit QLoRA, extending the adapters from attention projections to the MLP layers. A small classification head is trained with binary cross-entropy, and training ends with one extra epoch on the development set.

On the official test set the system scored 0.90, tying for fifth place, compared with 0.63 for a DeBERTa-v3-large baseline. The authors attribute most of the gain to model scale. They also show that task-specific fine-tuning clearly beats zero-shot prompting of a larger model, which matters for reasoning tasks with little training data.

Key points

  • Data and metric: 1,819 train, 400 dev and 612 test instances, covering 36, 8 and 12 topics. In training, 42.7% of instances have several correct answers. Scoring gives 1.0 for an exact match, 0.5 when the prediction contains some but not all of the correct answers and nothing incorrect, and 0 otherwise.
  • Retrieval settings:
    • BM25 weight α = 0.3.
    • About 300-character chunks with 50-character overlap.
    • Top-3 chunks retrieved.
    • Chunks containing extracted causal edges have their scores multiplied by 1.5.
  • Extended LoRA:
    • Seven target modules (q/k/v/o_proj, gate/up/down_proj).
    • 268M trainable parameters, which is 0.81% of about 33B.
    • Rank 32, α = 64.
    • Trained on a single H100 80GB in roughly 6–8 hours.
  • Results:
    • Best dev score: 0.9363.
    • Test scores across submissions: 0.63 (baseline), 0.86 (7B), 0.88 (32B), 0.90 (Extended LoRA).
    • A fine-tuned 7B model (0.86) beat a zero-shot 32B model (0.70).
  • Negative findings and errors:
    • Following 2-hop causal paths brought no gain.
    • Multi-agent debate dropped the score to 0.59.
    • 87.7% of test instances were matched exactly.
    • The "none of the others" option was always handled correctly.

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 worksComputer Vision and Pattern Recognition · 2,869 worksInformation Systems · 1,437 works

🧭 Topics 2Research subjects and languages the work is about.Reasoning · 1,426 worksInformation retrieval & search · 473 works

Cite

Cite as Duy Minh Dao Sy, Nguyen Tran, Trung Kiet Huynh, Phu Quy Nguyen Lam and Phu Hoa Pham. 2026. HCMUS RepeatedGames at SemEval-2026 Task 12: CausalRAG: Synergizing Causal Graph Retrieval and Extended LoRA for Abductive Reasoning. 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-1.59

@inproceedings{sy2026hcmus,
  title = {{HCMUS RepeatedGames at SemEval-2026 Task 12: CausalRAG: Synergizing Causal Graph Retrieval and Extended LoRA for Abductive Reasoning}},
  author = {Duy Minh Dao Sy, Nguyen Tran, Trung Kiet Huynh, Phu Quy Nguyen Lam and Phu Hoa Pham},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.59},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.59}
}

References

26 references, 11 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Chandra Bhagavatula, Ronan Le Bras, Chaitanya Malaviya, Keisuke Sakaguchi, Ari Holtzman, Hannah Rashkin, Doug Downey, Wen-tau Yih, and Yejin Choi. 2020. Abductive commonsense reasoning. In International Conference on Learning Representations.
  2. Duy Minh Dao-Sy, Trung Kiet Huynh, and 1 others. 2025a. JHARNA-MT: A copy-augmented hybrid of LoRA-tuned NLLB and lexical SMT with MBR decoding for low-resource indic languages. In Proceedings of the 1st Workshop on Multimodal Models for Low-Resource Contexts and Social Impact (MM-LoSo 2025). find
  3. Duy Minh Dao-Sy, Trung Kiet Huynh, and 1 others. 2025b. Leveraging lightweight entity extraction for scalable event-based image retrieval. arXiv preprint arXiv:2512.21221. — [arXiv:2512.21221](https://arxiv.org/abs/2512.21221) find
  4. Duy Minh Dao-Sy, Lam Phu Quy Nguyen, and 1 others. 2025c. DRAGON: Dual-encoder retrieval with guided ontology reasoning for medical normalization. In Proceedings of the 23rd Annual Workshop of the Australasian Language Technology Association (ALTA). find
  5. Tim Dettmers, Artidoro Pagnoni, Ari Holtzman, and Luke Zettlemoyer. 2023. Qlora: Efficient finetuning of quantized llms. Advances in Neural Information Processing Systems, 36.
  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 NAACL-HLT, pages 4171– 4186.
  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.
  8. Suchin Gururangan, Ana Marasović, Swabha Swayamdipta, Kyle Lo, Iz Beltagy, Doug Downey, and Noah A Smith. 2020. Don’t stop pretraining: Adapt language models to domains and tasks. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 8342–8360. find
  9. Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat, and Ming-Wei Chang. 2020. Realm: Retrieval-augmented language model pre-training. In International Conference on Machine Learning, pages 3929–3938.
  10. Pengcheng He, Xiaodong Liu, Jianfeng Gao, and Weizhu Chen. 2021. Deberta: Decoding-enhanced bert with disentangled attention. International Conference on Learning Representations.
  11. Neil Houlsby, Andrei Giurgiu, Stanislaw Jastrzebski, Bruna Morrone, Quentin De Laroussilhe, Andrea Gesmundo, Mona Attariyan, and Sylvain Gelly. 2019. Parameter-efficient transfer learning for nlp. In International Conference on Machine Learning, pages 2790–2799. find
  12. Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. Lora: Low-rank adaptation of large language models. International Conference on Learning Representations.
  13. Trung Kiet Huynh, Duy Minh Dao-Sy, and 1 others. 2025a. Systematic evaluation of machine learning and transformer-based methods for scientific telescope literature classification. In Proceedings of the Third Workshop for Artificial Intelligence for Scientific Publications (WASP 2025). find
  14. Trung Kiet Huynh, Duy Minh Dao-Sy, and 1 others. 2025b. Understanding LLM agent behaviours via game theory: Strategy recognition, biases and multi-agent dynamics. arXiv preprint arXiv:2512.07462. — [arXiv:2512.07462](https://arxiv.org/abs/2512.07462) find
  15. Trung Kiet Huynh, Duy Minh Dao-Sy, and 1 others. 2026. More at stake: How payoff and language shape LLM agent strategies in cooperation dilemmas. arXiv preprint arXiv:2601.19082. — [arXiv:2601.19082](https://arxiv.org/abs/2601.19082) find
  16. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, and 1 others. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. In Advances in Neural Information Processing Systems, volum…
  17. Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692.
  18. Lam Phu Quy Nguyen, Phu Hoa Pham, Duy Minh Dao-Sy, and 1 others. 2025. Beyond vision: Contextually enriched image captioning with multi-modal retrieval. arXiv preprint arXiv:2512.20042. — [arXiv:2512.20042](https://arxiv.org/abs/2512.20042) find
  19. SemEval-2026 Organizers. 2026. Semeval-2026 task 12: Abductive event reasoning. https://semeval.github.io/SemEval2026/tasks. ↗
  20. Charles Sanders Peirce. 1935. Collected Papers of Charles Sanders Peirce. Harvard University Press. find
  21. Pranav Rajpurkar, Jian Zhang, Konstantin Lopyrev, and Percy Liang. 2016. Squad: 100,000+ questions for machine comprehension of text. In Proceedings of EMNLP, pages 2383–2392.
  22. Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. In Proceedings of EMNLP-IJCNLP, pages 3982– 3992.
  23. Stephen Robertson and Hugo Zaragoza. 2009. The probabilistic relevance framework: Bm25 and beyond. Foundations and Trends in Information Retrieval, 3(4):333–389. find
  24. Qwen Team. 2024. Qwen2.5 technical report. arXiv preprint arXiv:2412.15115. find
  25. Miaoran Zhang, Marius Mosbach, David Adelani, Michael Hedderich, and Dietrich Klakow. 2022. Mcse: Multimodal contrastive learning of sentence embeddings. In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,… find
  26. Dawei Zhu, Xiaoyu Shen, Marius Mosbach, Andreas Stephan, and Dietrich Klakow. 2023. Weaker than you think: A critical look at weakly supervised learning. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 14229–14253. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sy-2026-hcmus-repeatedgames-semeval-2026-task/