Elamur Research⌕
☰
home / papers / 2026 / VVTE at ATE-IT: From Candidates to Terms: Hybrid Italian ATE with Dependency Heuristics, Gemini, and Random Forest Filtering

VVTE at ATE-IT: From Candidates to Terms: Hybrid Italian ATE with Dependency Heuristics, Gemini, and Random Forest Filtering

Paper · 2026 · Valentine G. L. Vandervoort and Giorgio Maria Di Nunzio

Summary

This paper presents a system built for the term extraction subtask (Subtask A) in the ATE-IT shared task held at EVALITA 2026. ATE-IT is the first large evaluation campaign for Italian automatic term extraction, and its data are institutional texts on municipal waste management. The domain is difficult because terms are sparse, Italian morphology is rich, and administrative terms are often long multi-word units or acronyms such as TARI and RAEE.

The pipeline has two phases:

  • Recall phase: candidates are pooled from spaCy (it_core_news_lg) dependency subtrees, noun chunks, acronym and pattern rules, and per-sentence zero-shot suggestions from Google Gemini.
  • Precision phase: candidates are cleaned and structurally filtered, then scored by a Random Forest using morphological, domain-keyword, positional and semantic-similarity features. A term is kept only if it is predicted positive with probability above 0.7.

On the test set, Type Precision reached 0.707, well above the baseline's 0.435, but recall and F1 fell below the baseline. The work shows how strict filtering in Italian ATE trades coverage for clean output and biases the system toward short terms.

Key points

  • Official test scores (system vs. baseline):
    • Micro P/R/F1: 0.364/0.473/0.411 vs. 0.497/0.559/0.526.
    • Type P/R/F1: 0.707/0.262/0.382 vs. 0.435/0.508/0.469.
  • The Random Forest weighted the positive class three times more (class_weight 1:3). It was selected because it was the most stable of the classifiers tried.
  • Adding the Grok LLM as an extra pre-filter after candidate extraction had a negligible effect on scores.
  • The main error was a "short-term bias": extracted terms rarely exceeded three words. The authors blame classifier rigidity and punctuation-based segmentation that splits long phrases.
  • Planned future work includes:
    • lowering the threshold and relaxing the filters;
    • better preservation of noun phrases;
    • transformer-based embeddings to capture rare, long-tail terms.

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 Valentine G. L. Vandervoort and Giorgio Maria Di Nunzio. 2026. VVTE at ATE-IT: From Candidates to Terms: Hybrid Italian ATE with Dependency Heuristics, Gemini, and Random Forest 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{vandervoort2026vvte,
  title = {{VVTE at ATE-IT: From Candidates to Terms: Hybrid Italian ATE with Dependency Heuristics, Gemini, and Random Forest Filtering}},
  author = {Valentine G. L. Vandervoort 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-1.58/}
}

References

5 references, 0 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. 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
  3. K. Kageura, B. Umino, Methods of automatic term recognition: A review, Terminology. International Journal of Theoretical and Applied Issues in Specialized Communication 3 (1996) 259–289. doi: 10. 1075/term.3.2.03kag . find
  4. K. Frantzi, S. Ananiadou, H. Mima, Automatic recognition of multi-word terms:. the C-value/NC-value method, International Journal on Digital Libraries 3 (2000) 115–130. doi: 10.1007/ s007999900023 . find
  5. Y. Chun, M. Kim, D. Kim, C. Park, H. Lim, Enhancing Automatic Term Extraction with Large Language Models via Syntactic Retrieval, in: W. Che, J. Nabende, E. Shutova, M. T. Pilehvar (Eds.), Findings of the Association for Computational Linguistics: ACL 2025, Association for Computational Linguistics,… find

Updated

This page in Markdown → https://md.research.elamur.ai/free/en/papers/vandervoort-2026-vvte-ate-it-candidates-terms/