Title: Peacemaker at ATE-IT: Leveraging Encoder Models for Automatic Term Extraction in Italian Waste Management Texts

Abstract

The significance of automatic term extraction has grown substantially within contemporary technology, with the technique now embedded in nearly every user-facing search engine. While recent progress has yielded encouraging outcomes for this field, achieving high accuracy remains challenging. These difficulties stem from issues such as the scarcity of annotated documents for training purposes and the complexity inherent in extracting multi-word expressions, particularly when domain shifts occur.

This paper introduces a cost-effective and interpretable approach to automatic term extraction, tailored specifically for Task A of the ATE Shared Task. The proposed method employs fine-tuning strategies designed to operate efficiently with minimal computational resources. To assess the system’s performance, we utilized both type-level and micro-level metrics, calculating precision, recall, and F1-scores to capture complementary dimensions of extraction quality.

Experimental data indicates that our approach delivers consistent and balanced results when compared to competing teams. Despite its relative simplicity, the technique offers a robust foundation for low-resource modeling. Ultimately, these results suggest that future enhancements—such as model expansion—can achieve higher performance levels while maintaining the interpretability that characterizes the current approach.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC