Commonsense Validation and Explanation in Arabic Text: A Comparative Study Using Arabic BERT Models
M Moneb Khaled, Aghyad AL Sayadi, Ashraf Elnagar · 2023
In today's rapidly digitizing world, the ability of machines to interpret and validate vast amounts of text, as humans do use commonsense knowledge, becomes crucial, especially in the fields of artificial intelligence and Natural Language Processing (NLP). This is particularly challenging for under-resourced, culturally rich languages like Arabic. To address this gap, this paper offers an in-depth evaluation of Arabic BERT models for commonsense validation and explanation tasks, utilizing a specially adapted Arabic version of the Commonsense Validation and Explanation (ComVE) dataset. We rigorously tested seven distinct Arabic BERT models, and our empirical results identify ARBERTv2 as the top-performing model. It achieves accuracy scores of 84.40% and 74.90% in commonsense validation and explanation tasks, respectively, thereby outperforming other state-of-the-art models in these domains. These findings not only underscore the efficacy of Arabic BERT models but also contribute to the broader initiative to extend machine commonsense reasoning capabilities to under-resourced languages.