Earticle

This paper examines whether artificial intelligence (AI) interventions in South Korea's Korean Sign Language (KSL) ecosystem will support language vitality or inadvertently accelerate endangerment. Through qualitative analysis of eight major government and private-sector AI projects (2024-2026), the study reveals a paradox: while technical capacity is rapidly maturing through national-scale datasets, commercial translation platforms, and institutional deployments, the dominant logic positions KSL primarily as a conversion layer for navigating Korean-dominant institutions rather than as an autonomous language requiring domain expansion. Drawing on theories of language endangerment, linguistic racism, and sociotechnical systems, the analysis demonstrates that current AI development optimizes institutional efficiency over linguistic vitality, risks imposing standardization through avatar rendering and constrained datasets, and lacks robust deaf community governance over objectives and evaluation criteria. These patterns are particularly consequential given KSL's structural vulnerabilities: disrupted intergenerational transmission, demographic collapse, and medicalized ideologies privileging cochlear implantation and speech outcomes. The findings suggest that without fundamental reorientation toward vitality-centered evaluation, deaf-led governance, and expansion of KSL-first domains, AI infrastructure may become another mechanism of technologically sophisticated assimilation rather than revitalization, making institutional navigation more efficient while KSL's own communicative life contracts.

The rapid integration of generative artificial intelligence into educational contexts has prompted increasing interest in its potential role in second language writing instruction. This study explores English as a Foreign Language (EFL) university learners’ perceptions of ChatGPT-based self-feedback in comparison with conventional offline teacher feedback, with particular attention to differences across learner proficiency levels. The participants consisted of first-year university students enrolled in an academic writing course, who were divided into a lower-proficiency group (n = 137) and an intermediate-proficiency group (n = 50). Employing a mixed-methods research design, the study examined learners’ overall perceptions of feedback, their preferences for feedback sources, and their prompt patterns of interaction with ChatGPT within a process-oriented writing framework. Quantitative survey data were complemented by qualitative analyses of learners’ written responses and the prompts they used when engaging with ChatGPT for self-feedback during multiple stages of writing revision. The results revealed that learners in both proficiency groups demonstrated generally positive attitudes toward both ChatGPT-based self-feedback and teacher-provided feedback. Across proficiency levels, the immediacy, accessibility, and convenience of ChatGPT were consistently identified as its most salient advantages, enabling learners to revise their writing efficiently and maintain sustained engagement throughout the writing process. However, notable proficiency-related differences emerged in feedback preferences. Intermediate-level learners exhibited a stronger preference for teacher feedback, emphasizing the value of instructors’ professional expertise, interpretive guidance, and discourse-level comments that extended beyond surface-level language issues. In contrast, lower-level learners expressed a preference for a combined feedback approach that integrated ChatGPT-based self-feedback with traditional teacher feedback. This preference reflects their need for complementary support, in which immediate AI-generated assistance helps address linguistic difficulties while teacher feedback provides authoritative validation and direction. Further analysis of ChatGPT interaction prompts indicated that learners primarily utilized the tool for local-level linguistic support, including grammar correction, vocabulary choice, and sentence-level refinement, rather than for higher-order writing concerns such as organization, argument development, or rhetorical effectiveness. Importantly, the findings also suggest that the pedagogical effectiveness of AI-based feedback is closely associated with learners’ metacognitive awareness and their ability to design effective prompts. Learners who demonstrated greater awareness of their writing needs and more strategic prompt use appeared to benefit more substantially from ChatGPT-based feedback. These results highlight the potential of AI-assisted feedback as a supplementary tool in EFL writing instruction, while underscoring the continued importance of teacher feedback and the need for explicit guidance in developing learners’ prompt-design and self-regulation skills.

This study proposes an interdisciplinary framework that integrates Visual Grammar with Artificial Intelligence (AI) to quantify multimodal semantics in sign language. Focusing on the lexical item “DANGER” in Korean Sign Language (KSL) and American Sign Language (ASL), we examine how representational, interactive, and compositional meanings are manifested in 3D kinematic patterns extracted from video. The analysis provides empirical evidence of contrasting representational strategies: KSL shows an internalized Z-axis “Impact” vector toward the chest, while ASL exhibits an externalized X-axis “Friction” vector associated with contact-based motion. In addition, AI-based quantification of non-manual markers reveals a consistent face–manual timing relationship, with facial tension preceding the manual sign by 185 ms (KSL) and 245 ms (ASL), supporting the theoretical view of facial expressions as grammatical anchors in multimodal communication. By moving beyond qualitative description toward measurement-based validation, this work positions AI as a methodological instrument for Digital Sign Humanities and demonstrates how micro-temporal and micro-spatial patterns—often difficult to detect through manual observation—can be systematically linked to linguistic interpretation. Future work will extend this framework to larger datasets and additional sign languages toward a global digital atlas of sign-language kinematics.