ATE-IT at EVALITA 2026: Overview of the Automatic Term Extraction Italian Testbed Task

Nicola Cirillo, Giorgio Di Nunzio, Federica Vezzani

Keywords: Automatic Term Extraction, Terminology, Italian NLP, Shared Task, EVALITA 2026

Abstract:

This paper presents an overview of the Automatic Term Extraction - Italian Testbed (ATE-IT) shared task, organised within the EVALITA 2026 evaluation campaign.
The task addresses the scarcity of benchmarks for Italian Automatic Term Extraction (ATE) by proposing a challenge focused on the domain of municipal waste management.
Participants were invited to tackle two subtasks: (A) Term Extraction, aiming to identify domain-specific terms in institutional texts, and (B) Term Variants Clustering, focusing on grouping morphological and semantic term variants.
Nine teams participated, submitting a total of 13 runs. The comparative analysis reveals that fine-tuned Transformer architectures generally outperform naive zero-shot Large Language Model (LLM) prompting, while hybrid approaches appear most effective for semantic clustering.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 3