This article reviews recent applications of deep learning to Arabic poetry between 2020 and 2025. This review follows a PRISMA-aligned systematic review design and synthesizes 45 peer-reviewed studies identified through major interdisciplinary databases and citation tracking. It examines three main clusters: meter detection, automatic poem generation and poetic interpretation. The review addresses a gap in current scholarship by showing that research on AI and Arabic poetry has largely measured formal performance while giving less attention to symbolic reasoning, cultural grounding and interpretive accountability. The findings show a clear task asymmetry. Models perform strongly on structural tasks such as meter detection and show moderate progress in metrically fluent generation. However, they tend to struggle with metaphor, historical awareness, cultural memory and symbolic continuity. These limitations are reported even in studies using large transformer-based systems, suggesting that scale alone does not appear to resolve interpretive difficulty. The article introduces Computational Arabic Poetic Intelligence (CAPI), which distinguishes structural, semantic and symbolic competence. CAPI is proposed as an evaluative framework rather than a descriptive taxonomy. The review argues that progress in cultural AI and digital humanities should be assessed through culturally grounded evaluation, not through accuracy or fluency metrics alone.
Alkhalaileh et al. (Wed,) studied this question.