The paper proposes a framework for evaluating zero-shot named entity recognition (NER) in Italian, which includes in-domain, out-of-domain, and unseen named entity scenarios.
The authors introduce SLIMER-IT, an Italian version of the SLIMER model, which uses instruction tuning and prompts enriched with definitions and guidelines to perform zero-shot NER. SLIMER-IT is evaluated against various state-of-the-art approaches, including token classification models and other zero-shot NER methods.
The results show that SLIMER-IT, particularly when using the LLaMAntino-3-ANITA backbone, significantly outperforms other models in the unseen named entity scenario, demonstrating the effectiveness of the definition and guideline-enriched prompts. The authors also explore the impact of different language model backbones on SLIMER-IT's performance.
The paper highlights the importance of addressing zero-shot NER, especially for languages like Italian where NER is understudied outside of traditional domains and entity types. The proposed evaluation framework and the SLIMER-IT approach contribute to advancing the state-of-the-art in zero-shot NER for the Italian language.
To Another Language
from source content
arxiv.org
Viktige innsikter hentet fra
by Andrew Zamai... klokken arxiv.org 09-25-2024
https://arxiv.org/pdf/2409.15933.pdfDypere Spørsmål