SLIMER-IT: Zero-Shot NER on Italian Language
Andrew Zamai, Leonardo Rigutini, Marco Maggini, Andrea Zugarini
2024-09-25
Summary
This paper introduces SLIMER-IT, a new system for Named Entity Recognition (NER) specifically designed for the Italian language. It uses a technique called zero-shot learning, allowing the model to identify named entities without needing extensive training on Italian data.
What's the problem?
Traditional NER systems often require a lot of labeled data to recognize names of people, places, and organizations in text. This is especially challenging for languages like Italian, where high-quality datasets are limited. Additionally, these systems struggle to recognize new types of entities that they haven't been trained on, making them less flexible and effective.
What's the solution?
To address these issues, the researchers developed SLIMER-IT, which leverages zero-shot learning. This means that instead of needing specific training on Italian data, SLIMER-IT uses knowledge from other languages and applies it to Italian. The system uses prompts that include definitions and guidelines to help it recognize entities in text. This approach allows SLIMER-IT to perform well even on types of entities it has never encountered before.
Why it matters?
This research is significant because it enhances the capabilities of AI in processing the Italian language without the need for large amounts of training data. By using zero-shot learning, SLIMER-IT can effectively identify named entities in various contexts, which is crucial for applications like information extraction and natural language understanding. This advancement could lead to better tools for working with under-resourced languages and improve accessibility in AI technologies.
Abstract
Traditional approaches to Named Entity Recognition (NER) frame the task into a BIO sequence labeling problem. Although these systems often excel in the downstream task at hand, they require extensive annotated data and struggle to generalize to out-of-distribution input domains and unseen entity types. On the contrary, Large Language Models (LLMs) have demonstrated strong zero-shot capabilities. While several works address Zero-Shot NER in English, little has been done in other languages. In this paper, we define an evaluation framework for Zero-Shot NER, applying it to the Italian language. Furthermore, we introduce SLIMER-IT, the Italian version of SLIMER, an instruction-tuning approach for zero-shot NER leveraging prompts enriched with definition and guidelines. Comparisons with other state-of-the-art models, demonstrate the superiority of SLIMER-IT on never-seen-before entity tags.