Elamur Research⌕
☰
home / papers / 2026 / TrietNLP at ATE-IT: A Hybrid Pipeline for Italian Waste Management Terminology Analysis

TrietNLP at ATE-IT: A Hybrid Pipeline for Italian Waste Management Terminology Analysis

Paper · 2026 · Nguyen Minh Triet and Dang Van Thin

Summary

The paper presents TrietNLP, a system the authors built to compete in EVALITA's ATE-IT 2025 shared task. The task targets terminology in Italian municipal solid waste management texts, such as regulations, notices and administrative acts. The two subtasks are handled by separate pipelines.

For term extraction (Subtask A), the authors frame the problem as BIO sequence tagging. Embeddings from xlm-roberta-large pass through a Linear+Tanh+LayerNorm projection block into a CRF, whose BIO-violating transitions are fixed to a large negative value. Nested gold terms are flattened with a Longest-Match-First rule.

For variant clustering (Subtask B), lightweight rules first form seed clusters. These rules strip accents and stopwords and apply light Italian stemming of vowel endings. Gemini 2.5 Flash then sorts the remaining terms in batches of 100. Each batch prompt includes every cluster built so far, so synonyms that land in distant batches still end up together. Terms the model invents are filtered out.

The system placed second in Subtask A and first in Subtask B. This suggests that rules combined with an LLM that carries context across batches work well for clustering domain synonyms and acronyms in Italian.

Key points

  • Data: 2,308 / 577 / 1,142 sentences (train / dev / test), with 2,218 and 451 annotated terms in train and dev.
  • Subtask A setup: XLM-RoBERTa Large + projection + CRF (PyTorch-CRF), Viterbi decoding, max length 256, batch 16, learning rate 1e-5, 6 epochs, dropout 0.2.
  • Subtask B official test results: BCubed P/R/F1 of 0.528/0.378/0.441, against 0.359 F1 for the runner-up and 0.245 for the baseline.
  • Prompt rules: acronyms are merged with their expansions (e.g., CCR with Centro Comunale di Raccolta), while hypernym–hyponym pairs are kept apart.
  • Reported errors: articles and prepositions leak into extracted spans (l'isola ecologica), very long terms get truncated, rifiuti is confused with its subtypes, and early clustering mistakes propagate through later batches.

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 2The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 works

Cite

Cite as Nguyen Minh Triet and Dang Van Thin. 2026. TrietNLP at ATE-IT: A Hybrid Pipeline for Italian Waste Management Terminology Analysis. Proceedings of the Ninth Evaluation Campaign of Natural Language Processing and Speech Tools for Italian. Final Workshop (EVALITA 2026).

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{triet2026trietnlp,
  title = {{TrietNLP at ATE-IT: A Hybrid Pipeline for Italian Waste Management Terminology Analysis}},
  author = {Nguyen Minh Triet and Dang {Van Thin}},
  booktitle = {Proceedings of the Ninth Evaluation Campaign of Natural Language Processing and Speech Tools for Italian. Final Workshop (EVALITA 2026)},
  year = {2026},
  url = {https://aclanthology.org/2026.evalita-1.57/}
}

References

7 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. N. Cirillo, G. M. Di Nunzio, F. Vezzani, ATE-IT at EVALITA 2026: Overview of the Automatic Term Extraction Italian Testbed Task, in: Proceedings of the Ninth Evaluation Campaign of Natural Language Processing and Speech Tools for Italian. Final Workshop (EVALITA 2026), CEUR.org, Bari, Italy, 2026. find
  2. A. Conneau, K. Khandelwal, N. Goyal, V. Chaudhary, G. Wenzek, F. Guzmán, E. Grave, M. Ott, L. Zettlemoyer, V. Stoyanov, Unsupervised cross-lingual representation learning at scale, in: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, Association for Computatio… find
  3. J. Lafferty, A. McCallum, F. C. Pereira, Conditional random fields: Probabilistic models for segmenting and labeling sequence data, in: Proceedings of the eighteenth international conference on machine learning (ICML), 2001, pp. 282–289. find
  4. G. Team, R. Anil, S. Borgeaud, Y. Wu, J.-B. Alayrac, J. Yu, R. Soricut, J. Schalkwyk, A. M. Dai, A. Hauth, et al., Gemini: a family of highly capable multimodal models, arXiv preprint arXiv:2312.11805 (2023).
  5. R. Verborgh, M. Röder, R. Usbeck, A.-C. Ngonga Ngomo, GERBIL – Benchmarking Named Entity Recognition and Linking Consistently, Semantic Web 9 (2018) 605–625. URL: https://doi.org/10. 3233/SW-170286. doi: 10.3233/SW-170286 . — [doi:10.3233/sw-170286](https://doi.org/10.3233/sw-170286) find
  6. A. Bagga, B. Baldwin, Entity-based cross-document coreferencing using the vector space model, in: Proceedings of the 36th Annual Meeting of the Association for Computational Linguistics and the 17th International Conference on Computational Linguistics (COLING-ACL’98), Montreal, 1998, pp. 79–85.
  7. E. Amigó, J. Gonzalo, J. Artiles, F. Verdejo, A Comparison of Extrinsic Clustering Evaluation Metrics Based on Formal Constraints, Information Retrieval 12 (2009) 461–486. find

Updated

This page in Markdown → https://md.research.elamur.ai/free/en/papers/triet-2026-trietnlp-ate-it-hybrid-pipeline/