Proceeding of the Faculty of technical Sciences, Novi Sad

Electrotechnical and Computer Engineering

Vol. 41 No. 08 (2026): Proceedings of the Faculty of Technical Sciences

Comparative Analysis of NER Models for Judgements in Montenegrin

  • Branislav Roljic
  • Stevan Gostojic
DOI:
https://doi.org/10.24867/
Submitted
September 7, 2026
Published
2026-09-09

Abstract

Named entity recognition in under-resourced languages is hindered by limited annotated data and the absence of domain-adapted models. This study evaluates transformer-based and generative NER approaches on a manually annotated corpus of Montenegrin legal texts. The methodology integrates linguistic insight with deep learning techniques to adapt models to the legal domain. Through comprehensive experiments, we examine performance across entity types, assess generalization, and identify key strengths and limitations of each method. The results support the development of NLP tools for Montenegrin legal language and highlight directions for advancing NER in low-resource legal settings. 

References

  1. [1] I. Keraghel, S. Morbieu, и M. Nadif, „Recent
  2. Advances in Named Entity Recognition: A
  3. Comprehensive Survey and Comparative Study“, 20.
  4. Децембар 2024., arXiv: arXiv:2401.10825. doi:
  5. 10.48550/arXiv.2401.10825.
  6. [2] H. Darji, J. Mitrović, и M. Granitzer, „German BERT
  7. Model for Legal Named Entity Recognition“, у
  8. Proceedings of the 15th International Conference on
  9. Agents and Artificial Intelligence, 2023, стр. 723–
  10. 728. doi: 10.5220/0011749400003393.
  11. [3] M. Bogdanović, M. Frtunić Gligorijević, J. Kocić, и
  12. L. Stoimenov, „An Analysis of the Training Data
  13. Impact for Domain-Adapted Tokenizer
  14. Performances—The Case of Serbian Legal Domain
  15. Adaptation“, Appl. Sci., том 15, изд. 13, стр. 7491,
  16. Јули 2025, doi: 10.3390/app15137491.
  17. [4] I. Ait Talghalit, H. Alami, и S. O. El Alaoui,
  18. „Exploring Different Annotation Schemes for Single
  19. and Consecutive Named Entity Recognition in the
  20. Arabic Biomedical Domain using Transformer
  21. Models and Contextual Semantic Embeddings“, Eng.
  22. Technol. Appl. Sci. Res., том 15, изд. 2, стр. 21854–
  23. 21860, Апр. 2025, doi: 10.48084/etasr.10019.
  24. [5] M. Škorić, „New Language Models for Serbian“,
  25. Infotheca, том 24, изд. 1, стр. 7–28, 2024, doi:
  26. 10.18485/infotheca.2024.24.1.1.
  27. [6] V. Kalušev и B. Brkljač, „Named entity recognition
  28. for Serbian legal documents: Design, methodology
  29. and dataset development“, 14. Фебруар 2025., arXiv:
  30. arXiv:2502.10582. doi: 10.48550/arXiv.2502.10582.
  31. [7] „Sudovi Crne Gore“. Приступљено: 28. Октобар
  32. 2025. [На Интернету]. Available at:
  33. https://sudovi.me/sdvi/odluke
  34. [8] Playwright. [На Интернету]. Приступљено: 28.
  35. Октобар 2025. Available at: https://playwright.dev/
  36. [9] LabelStudio. [На Интернету]. Приступљено: 28.
  37. Октобар 2025. Available at: https://labelstud.io/
  38. [10] seqeval. [На Интернету]. Приступљено: 28.
  39. Октобар 2025. Available at:
  40. https://huggingface.co/spaces/evaluate-metric/seqeval