LSTM-based deep learning model for named entity recognition in Yoruba language with events

Named entity recognition is a critical component of narrative understanding, and its challenges are amplified in underresourced languages such as Yoruba. This paper presents an approach to named entity recognition tailored to the Yoruba language, harnessing the power of Long Short-Term Memory (LSTM) networks. The proposed LSTM-based model is meticulously designed to capture not only named entities, but also event entities across various event types. The LSTM architecture, coupled with a Time Distributed Output layer, ensures precise temporal alignment and parallelization for accurate event trigger identification. In an experiment conducted on a Yoruba event extraction dataset, the model achieves an F1-score of 0.8967, a precision of 0.9391, and a recall of 0.8579 across various event types, showcasing its robustness and adaptability within the Yoruba linguistic context.