Event Nugget Detection using Pre-trained Language Models

This paper handles the task of event nugget detection. In fact, deep learning methods were able to manage the extraction of relevant learned features. However, these methods tend to rely on NLP-Toolkits, as they feed gradually handcrafted features into their initial model. To alleviate this dependency and offer a deeper semantic understanding of the information encompassed in data, we investigate the use of pre-trained language models. The proposed approach uses the RoBERTa model because it offers a robust context-sensitive and pertinent representation of trends in data. The results demonstrate that our approach significantly outperforms its BERT-based variants and state-of-the-art approaches.