Elamur Research⌕
☰
home / papers / 2026 / SMTE at ATE-IT: Ensemble Term Extraction with Italian BERT, spaCy, and Vocabulary-Based Filtering

SMTE at ATE-IT: Ensemble Term Extraction with Italian BERT, spaCy, and Vocabulary-Based Filtering

Paper · 2026 · Sofia Maule and Giorgio Maria Di Nunzio

Summary

The paper presents SMTE, a system for Subtask A of the ATE-IT shared task at EVALITA 2026. The task is to extract single- and multi-word domain terms from Italian institutional sentences about municipal waste management. Outputs must be lowercase, have no duplicates within a sentence, and exclude terms nested inside longer ones.

SMTE combines two supervised taggers:

  • dbmdz/bert-base-italian-uncased, fine-tuned with BIO labels;
  • a spaCy NER component (it_core_news_md), retrained for a single TERM label.

A whitelist of normalized gold terms from train+dev guides how their outputs are merged. Short BERT fragments are swapped for longer spaCy spans found in the vocabulary, and further vocabulary-backed spaCy multiwords are added. Rule-based cleanup then discards truncated spans, generic single words and nested terms.

The authors report that the ensemble beats either tagger alone, whereas LLM reranking did not raise Micro F1. The study shows that strict, deterministic constraint handling matters, because every rule violation counts as a false positive.

Key points

  • Evaluation uses Micro F1 (over occurrences) and Type F1 (over unique terms). The official scores are in a section missing from the extracted text.
  • BERT setup: HuggingFace Trainer on train+dev, learning rate 2×10⁻⁵, batch 16, 7 epochs, weight decay 0.01, warmup ratio 0.1, seed 42.
  • spaCy setup: 40 iterations, dropout 0.1, batch 8. Sentences without any TERM annotation are excluded from training.
  • Post-processing removes multiword candidates ending in a function word (e.g., "di", "dei", "e"). It resolves nesting by greedily keeping the longest character spans.
  • Main limitation: the gold-derived vocabulary may cap recall on unseen test terminology. Code is available at github.com/sofiamaule/ATE-IT_SofiaMaule.

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

🎓University of PaduaPadua · Italy18 works · 4,260 citations

 Fields, topics and concepts

📚 Shelves 4The 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 works

Cite

Cite as Sofia Maule and Giorgio Maria Di Nunzio. 2026. SMTE at ATE-IT: Ensemble Term Extraction with Italian BERT, spaCy, and Vocabulary-Based Filtering. 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{maule2026smte,
  title = {{SMTE at ATE-IT: Ensemble Term Extraction with Italian BERT, spaCy, and Vocabulary-Based Filtering}},
  author = {Sofia Maule and Giorgio Maria {Di Nunzio}},
  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-1.55/}
}

References

11 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. G. M. Di Nunzio, S. Marchesin, G. Silvello, A systematic review of Automatic Term Extraction: What happened in 2022?, Digital Scholarship in the Humanities 38 (2023) i41–i47. doi: 10.1093/ llc/fqad030 . find
  2. B. Jehangir, S. Radhakrishnan, R. Agarwal, A survey on Named Entity Recognition — datasets, tools, and methodologies, Natural Language Processing Journal 3 (2023) 100017. doi: 10.1016/j. nlp.2023.100017 . — [doi:10.1016/j](https://doi.org/10.1016/j) find
  3. F. Cutugno, A. Miaschi, A. P. Aprosio, G. Rambelli, L. Siciliani, M. A. Stranisci, Evalita 2026: Overview of the 9th evaluation campaign of natural language processing and speech tools for italian, in: Proceedings of the Ninth Evaluation Campaign of Natural Language Processing and Speech Tools for I… find
  4. 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
  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. doi: 10.3233/SW-170286 . — [doi:10.3233/sw-170286](https://doi.org/10.3233/sw-170286) find
  6. M. Martinelli, Ata_tutorship: Tutorship baseline code and reference implementations for ate-it, GitHub repository, 2025. URL: https://github.com/MMartinelli-hub/ATA_Tutorship. ↗
  7. J. Devlin, M.-W. Chang, K. Lee, K. Toutanova, Bert: Pre-training of deep bidirectional transformers for language understanding, in: Proceedings of NAACL-HLT, 2019, pp. 4171–4186. doi: 10.18653/ v1/N19-1423 .
  8. L. Ramshaw, M. Marcus, Text Chunking using Transformation-Based Learning, in: Third Workshop on Very Large Corpora, 1995.
  9. T. Wolf, L. Debut, V. Sanh, J. Chaumond, C. Delangue, A. Moi, P. Cistac, T. Rault, R. Louf, M. Funtowicz, J. Brew, Transformers: State-of-the-art natural language processing, in: Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, 2020, pp.…
  10. M. Honnibal, I. Montani, S. Van Landeghem, A. Boyd, spacy: Industrial-strength natural language processing in python, Website, 2025. URL: https://spacy.io/. ↗
  11. ATE-IT Shared Task, Ate-it results (evalita 2026) — official leaderboard, Website, 2025. URL: https://nicolacirillo.github.io/ate-it/results.html. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/maule-2026-smte-ate-it-ensemble-term/