Elamur Research⌕
☰
home / papers / 2026 / From Local Perspective to Global Reasoning: A Neuro-Symbolic Framework for Zero-Shot Relation Extraction

From Local Perspective to Global Reasoning: A Neuro-Symbolic Framework for Zero-Shot Relation Extraction

Paper · 2026 · Kailun Lyu, Fu Zhang, Zehan Li et al.

Summary

Zero-Shot Relation Extraction (ZSRE) requires labelling entity pairs in sentences with relations never seen during training. The authors argue that existing approaches (embedding similarity matching, synthetic data generation and LLM prompting) judge each sentence–pair instance on its own. As a result, they confuse unseen relations with close meanings, such as publisher vs. developer, when the wording is vague. The proposed framework, G-NSR, instead reasons jointly over a whole batch of candidate facts. Every sentence–relation combination becomes two textual statements, one affirming and one denying the relation. These are built with Duality Type-Constrained Relation Schemas, which also describe the expected entity types. A Proposition Encoder embeds the statements, and a Neuro-Symbolic Reasoner approximates implication and disjunction across the full set, so confidently judged propositions can correct uncertain ones. On Wiki-ZSL and FewRel, G-NSR beats prior systems in every evaluation setting and also has the highest inference throughput. The work shows that cross-instance logical consistency is a valuable signal for generalising to unseen relations.

Key points

  • Benchmarks: Wiki-ZSL and FewRel. G-NSR reports new state-of-the-art results in all settings, with an average gain of +4.14 over earlier methods.
  • Pipeline has three parts: proposition construction via Duality Type-Constrained Relation Schemas, a Proposition Encoder that fuses entity markers with relation features, and a three-stage Neuro-Symbolic Reasoner that learns an implication weight matrix among propositions.
  • Ablations show that raw relation labels already work reasonably well. Adding type constraints and duality (positive/negative pairs) gives consistent further gains.
  • Default configuration: M = 4 candidate relations during training and N = 32 sentences in both training and inference. Gains level off at larger values.
  • On FewRel with m = 5, 10 and 15 unseen labels, G-NSR processes more sentences per second than CE-DA, GLiREL and RE-GAR-AD. Code is available at github.com/LyuKaiLun/G-NSR.

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

🎓Northeastern UniversityShenyang · China42 works · 1,940 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

🧭 Topics 2Research subjects and languages the work is about.Reasoning · 1,426 worksInformation extraction · 167 works

Cite

Cite as Kailun Lyu, Fu Zhang, Zehan Li and Jingwei Cheng. 2026. From Local Perspective to Global Reasoning: A Neuro-Symbolic Framework for Zero-Shot Relation Extraction. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1492

@inproceedings{lyu2026local,
  title = {{From Local Perspective to Global Reasoning: A Neuro-Symbolic Framework for Zero-Shot Relation Extraction}},
  author = {Kailun Lyu, Fu Zhang, Zehan Li and Jingwei Cheng},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1492},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1492}
}

References

13 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Samy Badreddine, Artur d’Avila Garcez, Luciano Serafini, and Michael Spranger. 2022. Logic tensor networks. Artificial Intelligence, 303:103649.
  2. Antoine Bosselut, Ronan Le Bras, and Yejin Choi. 2021. Dynamic neuro-symbolic knowledge graph construction for zero-shot commonsense question answering. In Proceedings of the AAAI conference on Artificial Intelligence (AAAI), pages 4923–4931. find
  3. Jack Boylan, Chris Hokamp, and Demian Gholipour Ghalandari. 2025. GLiREL - generalist model for zero-shot relation extraction. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL), pages 823… find
  4. Chih-Yao Chen and Cheng-Te Li. 2021. ZS-BERT: Towards zero-shot relation extraction with attribute representation learning. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL), pages 3470–3479.
  5. Yew Ken Chia, Lidong Bing, Soujanya Poria, and Luo Si. 2022. RelationPrompt: Leveraging prompts to generate synthetic data for zero-shot relation triplet extraction. In Findings of the Association for Computational Linguistics (ACL), pages 45–57.
  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. Richard Evans and Edward Grefenstette. 2018. Learning explanatory rules from noisy data. Journal of Artificial Intelligence Research, 61:1–64. find
  8. Jiaying Gong and Hoda Eldardiry. 2024. Prompt-based zero-shot relation extraction with semantic knowledge augmentation. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (COLING), pages 13143–13156. find
  9. Xu Han, Hao Zhu, Pengfei Yu, Ziyun Wang, Yuan Yao, Zhiyuan Liu, and Maosong Sun. 2018. FewRel: A large-scale supervised few-shot relation classification dataset with state-of-the-art evaluation. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages… find
  10. Guozheng Li, Peng Wang, and Wenjun Ke. 2023. Revisiting large language models as zero-shot relation extractors. In Findings of the Association for Computational Linguistics (EMNLP), pages 6877–6892. find
  11. Guozheng Li, Peng Wang, Jiajun Liu, Yikai Guo, Ke Ji, Ziyu Shang, and Zijie Xu. 2024a. Meta in-context learning makes large language models better zero and few-shot relation extractors. In Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence (IJCAI), pages 6350–6… find
  12. Shilong Li, Ge Bai, Zhang Zhang, Ying Liu, Chenji Lu, Daichi Guo, Ruifang Liu, and Sun Yong. 2024b. Fusion makes perfection: An efficient multi-grained matching approach for zero-shot relation extraction. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Comp… find
  13. Tao Li and Vivek Srikumar. 2019. Augmenting neural networks with first-order logic. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (ACL), pages 292–302. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/lyu-2026-local-perspective-global-reasoning-neuro/