Abstract
Recognition of Arabic Sign Language (ARSL) remains a significant challenge due to the lack of extensive datasets, particularly those that reflect hand signs in real-life situations. The ArYSL Version 2 dataset was proposed to address such limitations by creating a nuanced Arabic Yemeni Sign Language to be used for Arabic sign translation tasks. The ArYSL Version 2 dataset is an expanded edition of Version 1, comprising 32 Arabic sign classes and 35,900 labeled RGB images collected from 35 participants of diverse ages and genders. In addition, ArYSL Version 2 dataset is enhanced with a curated dictionary of 357 Arabic words incorporating synonyms, dialectal variations and common misspellings. It is a suitable resource to utilize deep learning and fuzzy logic methods for bidirectional translation in both image-to-text and text-to-image tasks. The primary contribution of this work is the development and release of a large dynamic word-based, fully-labeled dataset of Arabic Yemeni Sign Language ArYSL Version 2 accompanied by details of Arabic data dictionaries. It is freely available online for the research community and can be publicly accessed at: https://doi.org/10.6084/m9.figshare.26114395.v1.