Development of an annotated Yoruba text corpus for automatic event extraction

Abstract The study documents the development of an annotated Yoruba text corpus which serves as a fundamental reference for training and evaluating automatic event extraction models specifically designed for the Yoruba language. While notable corpora like ACE and ERE have been established for high-resource and extensively annotated languages, African languages such as Yoruba have faced limitations in information extraction due to the lack of annotated datasets for training and model evaluation. In this study, the researchers took the initiative to preprocess raw Yoruba text obtained from selected Yoruba folktale books, ensuring the correct placement of tone marks, and proceeded to annotate the data with events and their corresponding arguments, including temporal and spatial information, utilizing the widely recognized BIO annotation format. This meticulously developed corpus now fills a critical void and can serve as a solid foundation for any future research endeavours involving the Yoruba language in the domain of information extraction.