강남욱 Kang Nam-wook

DOI:10.21716/TKFL.83.1 Vol.83(No.0) 1-42, 2026

This study offers a diachronic examination of in-service education (yeonsu) for teachers of Korean as a foreign language and, on that basis, proposes directions for its reconceptualization in the era of artificial intelligence. Departing from policy-centered periodizations, the study takes the practice of in-service education itself as its criterion and identifies three periods: the era of ‘training’, in which apprenticeship-style induction within institutions and the government-commissioned, Hangeul Society-operated training conferences for overseas Korean language teachers (1997-2010) formed a prototype of public-private partnership; the period in which ‘professional development’ was explored through university-based certificate programs and the state-commissioned Korean Language Globalization research projects; and the era of institutionalization under the Framework Act on the National Language (2005), characterized by a dual structure in which teacher certification was legislated while in-service education remained outside the statutory framework, sustained only by administrative programs. Two structural features―an orientation toward teachers abroad and provision through commissioned projects rather than statutory systems―are shown to run through this history. The study argues that AI, if designed as a tool for development rather than as a new form of training, can furnish the long-absent means for continuous, individualized in-service education, and proposes its reconceptualization from discrete events into an ongoing state of professional learning linked to certification.

한국어 교원 연수, in-service education for teachers of Korean as a foreign language, 교사 전문성 개발, teacher professional development, 한국어교육사, history of Korean language education, 재개념화, reconceptualization, 인공지능, artificial intelligence

권문화 Quan Wenhua , 장문정 Jang Moonjung

DOI:10.21716/TKFL.83.2 Vol.83(No.0) 43-72, 2026

This study aimed to explore the constituent factors and sub-components of AI literacy required of foreign learners of Korean translation and interpretation, and to develop basic assessment items for measuring this competency. Drawing on prior studies, five constituent factors-technical foundations, domain-specific performance, human-AI interaction, implementation, and ethical/societal aspects-were identified, from which 22 preliminary items were developed. Six experts in Korean language education, AI Education Expert, and translation studies then rated the items using Lynn’s (1986) Content Validity Index (CVI), evaluating each item’s relevance, clarity, representativeness, and measurability. The item-level content validity indices (I-CVI) ranged from .167 to 1.00, and the scale-level content validity index (S-CVI/Ave) was .871. Based on expert feedback, two items with low relevance were deleted or merged, and the remaining items were revised to be more accessible to foreign learners, resulting in a preliminary 20-item scale. This study contributes to the field by extending the discussion of AI literacy in translation and interpretation contexts to foreign learners of Korean and by establishing an empirical foundation for scale development through systematic expert content validation. Further research involving exploratory and confirmatory factor analyses with actual learner samples is warranted to refine and validate the scale

AI 리터러시, AI literacy, 한국어 통번역 교육, Korean translation and interpretation education, 외국인 한국어 학습자, KFL learners, 내용 타당도, content validity, 역량 척도, competency scale

김주은 Kim Jueun

DOI:10.21716/TKFL.83.3 Vol.83(No.0) 73-106, 2026

This study examines how Korean language teachers use generative AI in lesson preparation and how they perceive its effects on preparation time, instructional decision-making, and professional expertise. Written asynchronous interviews with eight in-service teachers, whose careers ranged from one to twenty years, were treated as multiple cases and analyzed thematically. Participants used AI to devise learner-appropriate grammar explanations and example sentences and to generate images, reading passages, test items, and slide materials, calibrating the specificity of their prompts to how far the output was to be used. Most treated it as a draft to be checked against textbooks and authoritative language resources before reconstructing it for their own learners. Perceptions of time savings varied widely and did not follow career length: one experienced teacher reported substantial savings only in a course requiring new test items each session, whereas one novice teacher used AI merely to obtain images, since item development lay outside her assigned duties. Cross-case comparison associated this variation with whether new materials had to be produced, whether institutional materials were judged sufficient, and the level of trust formed through early use. Most participants located their expertise in evaluating and adapting instructional content rather than in producing it, although two reported no change in this respect. As the data are self-reports gathered at a single point in time, the findings are offered as the perceptions of these eight teachers rather than as evidence of change in the professional knowledge of Korean language teachers as a whole.

한국어 교사, Korean language teacher, 인공지능, artificial intelligence, AI, 수업 준비, lesson preparation, 교사 전문성, teacher expertise, 다중 사례 연구, multi case study

이윤서 Lee Yoonseo , 노채환 Roh Chaehwan

DOI:10.21716/TKFL.83.4 Vol.83(No.0) 107-130, 2026

This study investigates the AI competencies of pre-service Korean language teachers within the Human-Centric Artificial Intelligence Pedagogy (HCAP) framework. A survey conducted with 61 pre-service teachers across five core HCAP domains: I-TK, I-CK, I-PK, HAIC-K, and Ethics-K. Overall perception was above average (M=3.42). Participants demonstrated strong human agency and ethical awareness, with the highest score recorded for teacher responsibility for AI-generated outputs (M=4.16). In contrast, lower scores were observed in the I-PK domain, particularly in the developing new Korean language assessment methods that include AI collaboration processes. Lower self-efficacy was also found in distinguishing how human teachers analyze the Korean language from how AI learns it statistically (M=2.89), as well as in implementing Universal Design for Learning (UDL) (M=2.90). These findings suggest teacher education must cultivate critical verification, process-oriented evaluation, classroom orchestration, and UDL-based instructional design beyond basic tool use.

HCAP, 예비 한국어 교사, Pre-service Korean teacher, 교사 지식, Teacher Knowledge, 인공지능 문해력, AI Literacy, 인공지능 활용 역량, AI Competency

명재권 Myoung Jaekwon

DOI:10.21716/TKFL.83.5 Vol.83(No.0) 131-198, 2026

This exploratory case study developed an AI agent for Korean writing assessment that incorporated a specific instructor’s assessment perspective and examined its performance on previously unseen student writing. Using undergraduate writing samples and assessment data from Instructor K, the study constructed a six-layer operational profile based on questionnaires, interviews, and authentic feedback. This profile was integrated with a case-based knowledge base through retrieval-augmented generation (RAG) to implement a multi-stage assessment agent. The data were divided into training, validation, and test sets at the student level. The agent generally identified the assessment domains and writing problems emphasized by Instructor K and reproduced key feedback practices, such as explaining problems, suggesting revisions, prompting self-monitoring, and guiding subsequent writing. Although its overall mean penalty score was similar to the instructor’s, discrepancies remained for individual texts and certain assessment domains. The findings indicate that the agent is better suited as an assistive tool providing preliminary evaluations and feedback informed by the instructor’s perspective than as an automated replacement, with the instructor retaining responsibility for final judgments.

Korean writing assessment, AI agent, instructor assessment perspective, Retrieval-Augmented Generation (RAG)

박승희 Park Seung-hee , 권태현 Kwon Tae-hyun , 정재은 Jung Jason J.

DOI:10.21716/TKFL.83.6 Vol.83(No.0) 199-245, 2026

Korean-language proficiency evaluation datasets for artificial intelligence systems require item-difficulty distributions designed to reliably differentiate capabilities across models. At the early stage of item development, however, response data are unavailable, and difficulty judgments therefore rely largely on qualitative criteria and expert experience. This study proposes the Design Difficulty Index (DDI), a framework for rating observable item-design features, together with their supporting rationale, prior to item administration. DDI distinguishes and connects, across the full assessment lifecycle, target difficulty, the pre-administration design score (s), the post-administration Item Response Theory (IRT) difficulty parameter (b), and the final operational level (L). The eight DDI Core rating items are reasoning depth, level of information integration, context and state dependence, Korean form and lexical knowledge requirements, pragmatic and social judgment requirements, external knowledge dependence, answer-structure complexity, and constraint-coordination complexity. These items are integrated with an assessment-element graph and quality gates. Three exploratory datasets were examined: NIA-53, a 53-item sample developed on the basis of expert experience for a National Information Society Agency (NIA) project proposal; Survey-65-4, comprising four writing items from the 65-item National Survey of Korean Language Proficiency conducted by the National Institute of Korean Language; and CSAT-945, comprising 945 items constructed from 21 Korean-language test forms actually administered in the College Scholastic Ability Test (CSAT) and its mock examinations, with 45 items taken by examinees in each administration. The DDI values in the original exploratory analyses remained either provisional single-analyst ratings or rule-based automated candidate ratings without independent review. In a separate independent re-rating of all 53 NIA items by two human raters, itemwise quadratic-weighted Cohen’s kappa ranged from -.002 to .485, exact agreement from 7.5% to 66.0%, and within-one-point agreement from 60.4% to 100.0%. This pattern provides preliminary reproducibility evidence for some items while identifying P, E, and X as priorities for boundary-rule refinement. NIA-53 showed a positive rank correlation between nominal levels and provisional unweighted design scores (ρ=.657, p<.001), although L2 and L3 overlapped substantially. Because nominal-level information was not concealed and NIA-53 was used to inspect the initial anchors, this association is interpreted as exploratory consistency within development data rather than independent convergent-validity evidence. In Survey-65-4, the intended difficulty explicitly specified by the item developers and the preliminary IRT category did not coincide for any of the four items, whereas the preliminary and main-survey IRT categories remained consistent for all four items. This case illustrates the potential value of treating intended difficulty and empirical difficulty as distinct layers and motivates an intermediate design layer for comparing and explaining their relationship. In CSAT-945, two-point and three-point items had mean provisional unweighted design scores of 21.51 and 35.10, respectively, with a large ordinal separation (Cliff’s δ=.699). Because official point values jointly reflect curricular importance, expected response time, and expected difficulty, the result is interpreted as exploratory convergence between automated candidate structural ratings and official point values. DDI contributes a computable record of item structure, quality status, evidence, and revision direction, together with a validation pathway linking pre-administration ratings to post-administration parameters. Further validation requires independent re-rating by three or more raters in external domains, explanatory models based on raw response data, external validation samples, and prospective item-revision experiments.

Design Difficulty Index, Pre-administration Difficulty Rating, Item Design, Item Response Theory, Artificial Intelligence Evaluation Dataset

박전규 Park Jeon Gue , 전형배 Jeon Hyung-bae , 임준호 Lim Joon-ho

DOI:10.21716/TKFL.83.7 Vol.83(No.0) 247-268, 2026

This study reports the design and performance of a Korean pronunciation error detection engine that identifies a learner’s first language (L1) and detects pronunciation errors by applying error rules specific to that language. The study used the Korean Speech Data of Foreign Speakers for AI Training (2021, 4,302 hours in total), released through AI Hub and collected from 2,000 foreign residents in Korea whose first languages were Vietnamese, Chinese, English, Japanese, or Thai. The engine was built in two stages. First, a self-supervised speech representation model was fine-tuned to identify five L1s; the model pre-trained on non-native Korean speech reached an accuracy of 96.1% on the test set, outperforming large multilingual models such as MMS and Whisper. Second, an error tagging scheme organized by onset, nucleus, and coda was established together with an L1-specific database of error rules, and detection was restricted to the target phonemes of the identified L1. Compared with a baseline in which no rules were applied, the error rate of pronunciation error detection decreased by an average of 60.9-70.6% for Vietnamese speakers and 56.2-65.2% for English speakers, and selecting rules by L1 consistently outperformed applying all rules at once. These findings show that L1 information can serve as effective prior knowledge for assessing non-native Korean pronunciation.

한국어 발음 교육, 모국어 식별, 발음 오류 검출, 모국어 간섭, 컴퓨터 보조발음 훈련(CAPT)

연준흠 Yeon Joonheum

DOI:10.21716/TKFL.83.8 Vol.83(No.0) 269-299, 2026

This study reports the design and development of a web-based self-monitoring tool for Korean ‘ㄹ’ pronunciation. Difficulty with ‘ㄹ’ is not confined to a single L1 background: it arises from the articulatory complexity of liquids and is compounded by L1 interference. Adopting a design and development research approach, the study first derived five design requirements from the phonetic characteristics of ‘ㄹ’, the hierarchical procedures of articulation therapy, and the self-monitoring principle of motor skill learning: pronunciation-environment based learning units, limited expansion from words to longer utterances, listening before production, model-self recording comparison with checklists, and self-assessment without automatic scoring. These requirements were implemented in a fixed web prototype containing 84 recorded model items across seven pronunciation environments, developed by the teacher-researcher with generative AI assisting implementation under the teacher’s pedagogical decisions. Implementation was then examined through static verification of the code and eight requirement-based test cases. The tests confirmed that the model audio, the fixed three-step learning sequence, and the absence of scoring code operated as intended, while revealing a usability defect and an unsuitable contrast item, both subsequently corrected. Learning effects and cross-L1 applicability are not claimed; they remain hypotheses for subsequent validation. (Seoul National University)

‘ㄹ’ 발음, Korean ‘ㄹ’ pronunciation, 조음 치료, articulation therapy, 자기 모니터링, self-monitoring, 발음 환경, pronunciation environment, 자가 점검 웹 도구, web-based self-monitoring tool

김유미 Kim Yumi , 김강희 Kim Kanghee , 최지희 Choi Jihee

DOI:10.21716/TKFL.83.9 Vol.83(No.0) 301-334, 2026

This study explores design directions for an AI-based Korean language learning platform in response to the increasing use of generative AI in Korean language education. In-depth interviews were conducted with 15 Korean language teachers, and the data were analyzed using thematic analysis. The findings indicate that teachers view AI as a complementary tool rather than a replacement for teachers. AI is expected to support repetitive practice, error detection, personalized learning, preliminary feedback, and data analysis, while teachers are expected to provide cultural and pragmatic guidance, emotional support, and final educational judgment. The results also suggest that AI-based learning should follow a cyclical process of diagnosis, personalization, performance, feedback, re-learning, and assessment. Teachers further emphasized the need to connect learning data with teacher dashboards and to integrate online AI-based learning with offline instruction. Based on these findings, the study proposes three core design principles: teacher-led and AI-assisted complementary learning, diagnosis-based cyclical personalized learning, and data-driven teacher integration within online-offline learning cycles. The study conceptualizes AI-based Korean language learning as a collaborative system connecting learners, AI, and teachers.

AI 기반 한국어 학습 플랫폼, AI-based Korean language learning platform, 한국어 교원, Korean language teachers, 심층 인터뷰, In-depth interviews, 교원-AI 협력, Teacher-AI collaboration, 주제 분석, Thematic analysis

김성숙 Kim Sung Sook , 이희경 Lee Hee-kyung , 정여훈 Jeong Yeohoon

DOI:10.21716/TKFL.83.10 Vol.83(No.0) 335-362, 2026

This study reports the first-year findings of a three-year project aimed at developing a generative AI-based Korean speaking assessment tool and evaluation system. The study begins with a review of prior research on AI-assisted Korean speaking education and assessment, identifying key achievements and limitations, and outlining the methodological framework adopted. To train AI for automatic item generation, a set of 20 speaking test sets were created across six proficiency levels. The research further examines the item development process, as well as the technical requirements of the online assessment platform and scoring system. Scoring criteria were established and applied in a preliminary evaluation with 12 learners and a main evaluation with 120 learners. Professional raters conducted two rounds of scoring, and their findings inform guidelines for large-scale item development and assessment implementation in subsequent phases. This study contributes to the advancement of AI-driven language assessment by integrating cultural relevance, adaptive learning, and rigorous evaluation standards.

인공지능, AI, 한국어, Korean, 말하기, speaking, 평가 도구, assessment tool, 평가 운영, implementation of assessment