<![CDATA[This article presents a critical narrative review of Scopus-indexed deep learning applications in Indonesian language education, organized around three pedagogically significant domains: language anxiety, literacy development, and language assessment. The study examined 134 Scopus-indexed publications (2022–2026) that applied deep learning techniques, including convolutional neural networks, recurrent architectures, Transformer-based models such as IndoBERT, and multimodal vision–language systems, to Indonesian and regional languages in Indonesia. Through the lens of second-language acquisition (SLA) and learner-centered pedagogy, the review synthesized empirical evidence across the three domains, identified recurring pedagogical potentials (early affective detection, inclusive literacy access, and real-time formative feedback), and highlighted persistent challenges, including construct reductionism, algorithmic bias against non-standard language varieties, data privacy concerns, and academic integrity issues. The findings revealed that although deep learning research on Indonesian languages has advanced rapidly from a technical perspective, its pedagogical foundation remains underdeveloped, as most studies prioritize engineering performance metrics rather than SLA-oriented learning outcomes. This article proposes an SLA-informed evaluation framework encompassing input quality, interaction authenticity, output opportunities, affective filters, and learner agency, while articulating a learner-centered, multilingual, and ethically governed agenda for AI-mediated language education in Indonesia. The findings suggest that the future of language education in Indonesia will depend not only on model sophistication but also on the pedagogical and ethical frameworks that guide their implementation. Quantitative analysis of the 134 studies revealed three key findings: (1) 89% of the reviewed studies reported only engineering metrics without addressing learning outcomes; (2) none of the studies reported formal ethics review procedures or subgroup validity analyses for non-standard language varieties; and (3) although 23 studies demonstrated technical capabilities applicable to language learning, only 7 positioned their work within SLA or pedagogical frameworks.]]>
Copyrights © 2026
