<![CDATA[Recent advances in Arabic Natural Language Processing (NLP) have been largely driven by transformer-based architectures and Large Language Models (LLMs), which outperform traditional approaches across tasks such as sentiment analysis, machine translation, text classification, and question answering. Nevertheless, Arabic NLP research remains predominantly technocentric and insufficiently connected to Second Language Acquisition (SLA) theories and pedagogical frameworks, thereby limiting its educational applicability. This study aims to systematically synthesize recent developments in LLM-based Arabic NLP and critically examine their potential integration into Arabic language learning, particularly within SLA and Computer-Assisted Language Learning (CALL) frameworks. A Systematic Literature Review guided by the PRISMA framework was conducted using Scopus-indexed studies published between 2025 and 2026. The selected studies were analyzed through thematic synthesis across five dimensions: model architecture, task domain, data strategies, linguistic focus, and pedagogical relevance. The findings demonstrate the growing dominance of LLMs, including AraBERT, AraGPT2, and ALLAM, alongside increased reliance on data augmentation, synthetic corpus generation, and dialect-specific adaptation. Evaluation practices are also beginning to extend beyond conventional performance metrics toward trustworthiness, safety, reasoning, and robustness. However, the review identifies a near absence of SLA-informed applications, Arabic LLM-based CALL systems, and human-centered evaluation involving usability, learner engagement, trust, and cognitive load. Despite this gap, LLMs offer substantial pedagogical potential through adaptive feedback, contextualized input, interactive dialogue, and personalized learning support. The study concludes that Arabic NLP requires stronger interdisciplinary integration with SLA and CALL to transform technological capabilities into learner-centered, pedagogically grounded, and empirically validated language-learning applications.]]>
