Elamur Research⌕
☰
home / papers / 2026 / Do Syntactic Features Help Biomedical Relation Extraction? An Empirical Study of Verb Token and Dependency Graph Augmentation

Do Syntactic Features Help Biomedical Relation Extraction? An Empirical Study of Verb Token and Dependency Graph Augmentation

Paper · 2026 · Mustafa Sikder and Ernest Kwegyir-Afful

Summary

The study tests whether explicit syntactic information improves transformer-based biomedical relation extraction beyond a standard typed entity marker setup. The baseline is BiomedBERT with PURE-style typed markers, pooling the start-marker states. The authors compare it with two extensions. The first appends the contextual vector of the dependency ROOT verb, identified with spaCy en_core_web_sm, to the entity representations. The second is a two-layer GCN that refines word-level hidden states over the full undirected dependency parse before the entity vectors are taken. They evaluate on ChemProt, the DDI corpus and AIMed (fold 1 only), with three seeds each. Neither extension beats the baseline consistently, and no difference reaches statistical significance. The GCN brings small gains on AIMed and ChemProt but hurts DDI, while the verb feature helps only on AIMed. A syntactic profile shows that DDI is far more passive. The authors argue its root verbs are often reporting predicates rather than relational ones. The paper is a useful cautionary result: syntax is not universally helpful, and a corpus's writing style should be checked before adopting such features.

Key points

  • GCN vs. baseline (micro F1): +0.007 on AIMed, +0.003 on ChemProt and −0.013 on DDI. Verb token: +0.004 on AIMed, −0.005 on ChemProt and −0.018 on DDI. All p > 0.05.
  • Passive voice in relation-bearing sentences is 50.7% for DDI, 27.0% for AIMed and 30.9% for ChemProt. Sentences with an active ROOT verb make up 45.0%, 58.6% and 55.3% respectively.
  • No usable ROOT verb was found for 10.3% (ChemProt), 11.7% (AIMed) and 7.8% (DDI) of relations. In those cases a learned NO_VERB_TOKEN embedding was used instead.
  • Training setup: AdamW, learning rate 2e-5, batch size 16 and seeds 0, 1 and 42. Training lasted 10 epochs for AIMed, while each of the remaining datasets was trained for 5 epochs.
  • Acknowledged limitations include the general-domain parser (no biomedical treebank), a single encoder and three seeds only. The code and converted datasets are released.

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

Also on the byline: United States Food and Drug Administration.

 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 worksBiochemistry, Genetics and Molecular Biology · 234 worksMolecular Biology · 231 works

🧭 Topics 1Research subjects and languages the work is about.Information extraction · 167 works

Cite

Cite as Mustafa Sikder and Ernest Kwegyir-Afful. 2026. Do Syntactic Features Help Biomedical Relation Extraction? An Empirical Study of Verb Token and Dependency Graph Augmentation. BioNLP 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.bionlp-1.12

@inproceedings{sikder2026do,
  title = {{Do Syntactic Features Help Biomedical Relation Extraction? An Empirical Study of Verb Token and Dependency Graph Augmentation}},
  author = {Mustafa Sikder and Ernest Kwegyir-Afful},
  booktitle = {BioNLP 2026},
  year = {2026},
  doi = {10.18653/v1/2026.bionlp-1.12},
  url = {https://doi.org/10.18653/v1/2026.bionlp-1.12}
}

References

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

  1. Sapna Bawankar and Sanjay Kumar. 2026. Scalable knowledge extraction from biomedical literature for drug discovery applications. Procedia Computer Science, 275:226–233. 7th International Conference on AI in Computational Linguistics. find
  2. Razvan Bunescu, Ruifang Ge, Rohit J. Kate, Edward M. Marcotte, Raymond J. Mooney, Arun K. Ramani, and Yuk Wah Wong. 2005. Comparative experiments on learning information extractors for proteins and their interactions. Artificial Intelligence in Medicine, 33(2):139–155. Information Extraction and Sum… find
  3. Yu Gu, Robert Tinn, Hao Cheng, Michael Lucas, Naoto Usuyama, Xiaodong Liu, Tristan Naumann, Jianfeng Gao, and Hoifung Poon. 2021. Domain-specific language model pretraining for biomedical natural language processing. ACM Trans. Comput. Healthcare, 3(1).
  4. María Herrero-Zazo, Isabel Segura-Bedmar, Paloma Martínez, and Thierry Declerck. 2013. The ddi corpus: An annotated corpus with pharmacological substances and drug–drug interactions. Journal of Biomedical Informatics, 46(5):914–920. find
  5. Matthew Honnibal, Ines Montani, Sofie Van Lan-deghem, and Adriane Boyd. 2020. spaCy: Industrial-strength Natural Language Processing in Python. https://github.com/mustafasikder/ ↗
  6. Biomedical-Relation-Extraction Ming-Siang Huang, Jen-Chieh Han, Pei-Yen Lin, Yu-Ting You, Richard Tzong-Han Tsai, and Wen-Lian Hsu. 2024. Surveying biomedical relation extraction: a critical examination of current datasets and the proposal of a new resource. Briefings in Bioinformatics, 25(3):bbae13… find
  7. Thomas N. Kipf and Max Welling. 2016. Semi-supervised classification with graph convolutional networks. CoRR, abs/1609.02907.
  8. Rabal-O. Lourenço A. Krallinger, M. 2017. Overview of the biocreative vi chemical-protein interaction track. Proceedings of the BioCreative VI Workshop,, 141-146. find
  9. Jinhyuk Lee, Wonjin Yoon, Sungdong Kim, Donghyeon Kim, Sunkyu Kim, Chan Ho So, and Jaewoo Kang. 2019. Biobert: a pre-trained biomedical language representation model for biomedical text mining. Bioinformatics, 36(4):1234–1240. trees improves relation extraction. In Proceedings of the 2018 Conference…
  10. Zexuan Zhong and Danqi Chen. 2021. A frustratingly easy approach for entity and relation extraction. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 50–61, Online. Association for Computational L…
  11. Min Zuo, Baoyu Zhang, Qingchuan Zhang, Wenjing Yan, and Dongmei Ai. 2022. An entity relation extraction method for few-shot learning on the food health and safety domain. Computational Intelligence and Neuroscience, 2022:1–11. find
  12. Sumit Madan, Justyna Szostak, Ravikumar Komandur Elayavilli, Richard Tzong-Han Tsai, Mehdi Ali, Longhua Qian, Majid Rastegar-Mojarad, Julia Hoeng, and Juliane Fluck. 2019. The extraction of complex relationships and their conversion to biological expression language (bel) overview of the biocreative… find
  13. Olga Majewska, Charlotte Collins, Simon Baker, Jari Björne, Susan Windisch Brown, Anna Korhonen, and Martha Palmer. 2021. Bioverbnet: a large semantic-syntactic classification of verbs in biomedicine. Journal of Biomedical Semantics, 12(1). find
  14. Tsendsuren Munkhdalai, Feifan Liu, and Hong Yu. 2018. Clinical relation extraction toward drug safety surveillance using electronic health record narratives: Classical learning versus deep learning. JMIR Public Health and Surveillance, 4(2):e29. find
  15. Livio Baldini Soares, Nicholas FitzGerald, Jeffrey Ling, and Tom Kwiatkowski. 2019. Matching the blanks: Distributional similarity for relation learning. CoRR, abs/1906.03158.
  16. Deming Ye, Yankai Lin, Peng Li, and Maosong Sun. 2022. Packed levitated marker for entity and relation extraction. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 4904–4917, Dublin, Ireland. Association for Computational Lingu…
  17. Yuan Zhang, Xin Sui, Feng Pan, Kaixian Yu, Keqiao Li, Shubo Tian, Arslan Erdengasileng, Qing Han, Wanjing Wang, Jianan Wang, Jian Wang, Donghu Sun, Henry Chung, Jun Zhou, Eric Zhou, Ben Lee, Peili Zhang, Xing Qiu, Tingting Zhao, and Jinfeng Zhang. 2025. A comprehensive large-scale biomedical knowled… find
  18. Yuhao Zhang, Peng Qi, and Christopher D. Manning. 2018. Graph convolution over pruned dependency find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sikder-2026-do-syntactic-features-help-biomedical/