Jiun Baek

DOI:10.15702/mall.2026.29.3.1 Vol.29(No.3) 1-23, 2026

Speaking development has long been challenging in EFL instruction due to limited opportunities for practice and exposure outside the classroom. This study compared three instructional conditions for EFL speaking in South Korea: traditional speaking instruction, MABL-based storytelling, and ChatGPT-supported MABL storytelling. One hundred lowerintermediate undergraduates in six intact class sections participated in this 16-week quasi-experimental study. Speaking performance was assessed through pre- and post-test oral proficiency interviews rated by two raters using a fourdimensional rubric (pronunciation, grammatical accuracy, fluency and coherence, vocabulary). Paired-samples t-tests showed all three groups improved significantly, with the largest effect size in the ChatGPT-supported group. A mixed ANOVA showed a non-significant main effect of group but a significant Time × Group interaction, indicating differing magnitudes of improvement across conditions. Follow-up gain-score analyses confirmed the ChatGPT-supported group improved significantly more than both the Control and MABL groups, while Control and MABL did not differ significantly. These findings suggest that integrating ChatGPT into a MABL storytelling framework may be associated with greater speaking gains than traditional instruction, though the quasi-experimental design and lack of usage data preclude attributing the improvement to ChatGPT alone. The absence of a MABL-Control difference is discussed in relation to learners’ mobile practices.

ChatGPT, EFL speaking, storytelling, blended learning, MABL Applicable level, higher education

Eun Seon Chung

DOI:10.15702/mall.2026.29.3.24 Vol.29(No.3) 24-39, 2026

With rapid advances in machine translation (MT), post-editing is now a common practice in the translation industry as well as language learning. As MT continues to improve in fluency and accuracy, the challenge of post-editing has shifted from identifying and correcting errors to making nuanced interpretive and pragmatic choices. This exploratory study examines how 12 advanced Korean EFL learners revise ChatGPT-generated English-to-Korean translations during light post-editing (LPE) and full post-editing (FPE), with particular attention to the types of decisions they make throughout the post-editing process. For each learner, the number of words changed in each task was counted, and the revisions were analyzed together with the learners’ written reflections. The learners made minimal changes in LPE but in FPE reworked the output into natural, audience-appropriate Korean through recurring types of revision in register, tone, person reference, sentence structure, word choice, and idiomatic expressions. Students’ written reflections suggested that they view MT as a resource whose output should be evaluated against one’s own linguistic and cultural judgment. Pedagogical implications for MT post-editing instruction in language education are discussed.

machine translation post-editing, light and full post-editing, AI-generated translation, English-to-Korean translation, advanced EFL learners

한은지 Eunji Han , 황현우 Hyunwoo Hwang

DOI:10.15702/mall.2026.29.3.40 Vol.29(No.3) 40-63, 2026

This study examined the instructional design principles and actual implementation of a five-stage gamified English class model built around Quizizz, using Gagné's Nine Events of Instruction as the analytical framework. A qualitative case study was conducted in a 12-week TOEIC course for 102 nursing students at a college in G Metropolitan City in 2024, using multiple instructor-generated data sources, including lesson plans, item design documents, and the instructor’s observational records. The results showed that the five-stage instructional model was systematically aligned with Gagné's nine events. This alignment revealed three key concepts: a 'ritual transition device,' a 'data-driven feedback loop,' and a 'dual motivational response structure.' The instructor’s observations also revealed a dual pattern of active engagement and competitive stress during the gamified stages. These findings suggest that effective gamification-based instruction depends on theoretically grounded instructional design rather than on tool use alone. Because learner responses in this study were documented exclusively through the instructor's observations rather than through direct learner accounts, the findings should be understood as instructor-mediated interpretations of learner engagement. This underscores the instructor’s role as an instructional designer in realizing the educational potential of gamification and offers practical guidance for digital English education.

gamification, Gagné, s Nine Events of Instruction, Quizizz, instructional design, EFL Applicable level, tertiary education

Hyangeun Ji , Hye Won Shin

DOI:10.15702/mall.2026.29.3.64 Vol.29(No.3) 64-84, 2026

Augmented reality (AR) has shown the potential to create new opportunities for constructive learning in second language classrooms. While educational researchers are increasingly recognizing this potential, there is currently a limited understanding of exactly how AR facilitates language learning, particularly with regard to second language vocabulary development in adolescent learners. To address this gap, a randomized controlled trial was conducted to examine whether augmented reality instruction could improve students' receptive vocabulary scores. Fifty Korean-speaking children (mean age = 10.4) were randomly assigned to one of three conditions: picture-based (PI) instruction, AR-based instruction, or a business-as-usual control condition. A linear mixed-effects model revealed that the vocabulary gains of the PI and AR groups from pretest to delayed posttest were significantly different over time. While post-hoc analysis revealed no differences in performance between the PI and AR groups, the effects of AR were statistically significant when compared to the business-as-usual control condition. These findings contribute to the advancement of vocabulary research by demonstrating the successful implementation of AR technology in an English as a foreign language classroom setting.

augmented reality, picture-based instruction, vocabulary learning, second language acquisition Applicable level, primary education

Moojoong Kim , Myunghwan Hwang

DOI:10.15702/mall.2026.29.3.85 Vol.29(No.3) 85-116, 2026

This study examined whether ChatGPT-based image-generation activities affect university EFL learners’ descriptive writing performance and how learners experience such activities. Forty undergraduate EFL learners at a Korean University were assigned to an experimental group, which generated images from English prompts and revised those prompts before writing, or a comparison group completing conventional pre-writing tasks. Both groups produced descriptive compositions before and after a four-session treatment. Writing samples were rated across six TEEP domains and analyzed using MANCOVA and Cohen’s f effect sizes, while experimental-group reflection logs were analyzed thematically. The results showed that the multivariate group effect did not reach statistical significance, although the associated effect size was large. Exploratory univariate analyses showed consistent medium-to-large effects favoring the experimental group in text organization, grammatical appropriateness, lexical appropriateness, and spelling. The thematic analysis indicated that learners experienced expanded linguistic resources, more concrete ideas, greater support for compositional organization, and enhanced metacognitive reflection, while also encountering technical instability, the cognitive burden of composing English prompts, concerns about overreliance, and occasional emotional strain. Overall, the study offers preliminary, exploratory evidence that the recursive prompt-image-revision cycle can direct learners’ attention to form-meaning relationships and push them toward more precise expression.

image-generation AI, descriptive writing, EFL writing, prompt literacy, multimodal writing

Jiyoung Lee , Yujong Park

DOI:10.15702/mall.2026.29.3.117 Vol.29(No.3) 117-141, 2026

This study examines corrective feedback (CF) displayed by a commercial AI-based speaking application, Speak, during Korean adult EFL learners’ spoken interaction. AI speaking applications are increasingly promoted as tools for personalized speaking practice, but less is known about how their feedback operates turn by turn or whether it functions as mediation for learner meaning-making and self-repair. Drawing on Lyster and Ranta’s (1997) CF taxonomy and a sociocultural-theoryinformed view of mediation, the study analyzed on-screen feedback records, audio recordings, and transcripts from 42 speaking sessions completed by six Korean adult learners over seven consecutive days, with post-use questionnaire responses used as supplementary data. Of the 622 on-screen CF instances identified, all met the study’s operational criterion for explicit correction, and none took the form of recast, clarification request, elicitation, repetition, or metalinguistic feedback without a corrected form. Although this feedback often supplied clear corrected forms with brief explanations, observable opportunities for uptake, self-repair, or negotiation were rare. Automatic speech recognition errors and semantic or contextual misinterpretations sometimes displaced learners’ intended meanings, making grammatically plausible corrections interactionally misaligned. Some episodes nonetheless showed localized mediational affordances. The study argues that the provision of corrective feedback should not be equated with mediation in AI-mediated speaking interaction.

corrective feedback, AI-mediated speaking, sociocultural mediation, meaning-making, self-repair

문경희 Kyunghee Moon , 김소연 Soyeon Kim , 최윤정 Yunjung Choi

DOI:10.15702/mall.2026.29.3.142 Vol.29(No.3) 142-176, 2026

This quasi-experimental study examined the effects of Self-Determination Theory (SDT)-based learning routines on upperelementary students’ English literacy development in a TED-Ed environment. Forty fifth- and sixth-graders from two Seoul elementary schools were assigned at the school level to an SDT group and a comparison group that used the same videos and instructional time; 37 students were analyzed. Over three weeks (15 sessions, 450 minutes), the SDT group received an integrated need-supportive routine combining choices of videos and tasks, scaffolded difficulty with process-focused feedback, and teacher and peer interactions; the comparison group used common materials, uniform tasks, and teacherdirected guidance. Outcomes included basic psychological need satisfaction, perceived quality of learning routines, and an English literacy test assessing reading and writing. Welch t tests showed that the SDT group reported greater gains in autonomy, competence, relatedness, and SDT scores and achieved larger literacy improvements than the comparison group. Findings suggest media-based English outcomes may depend not only on digital content but also on the motivational and self-regulatory structure through which learners repeatedly engage with it. Given the school-level assignment and small sample, findings should be interpreted as differential gains associated with SDT-based learning routines rather than definitive causal effects.

SDT, English literacy, learning routine, autonomy, motivation

Shinjae Park

DOI:10.15702/mall.2026.29.3.177 Vol.29(No.3) 177-204, 2026

AI-based writing feedback has become increasingly prevalent in L2 writing instruction, yet its effectiveness appears to vary across learning contexts and instructional conditions. This meta-analysis investigated the effectiveness of AI-based English writing feedback in L2 writing and examined differences in the effectiveness of AI-based English writing feedback across contextual and pedagogical conditions. A systematic search of SCOPUS and Web of Science identified 71 potentially relevant studies published between 2023 and 2025. Following a screening process based on predefined inclusion and exclusion criteria, eligible studies were synthesized using meta-analytic procedures. Effect sizes were calculated using Hedges’ g, and moderator analyses were conducted. The analysis revealed a large and statistically significant overall effect of AI-based feedback on L2 writing performance despite substantial heterogeneity across studies. Automated feedback produced larger effect sizes than mixed feedback combining AI with teacher or peer input, although both approaches yielded significant benefits. AI-based feedback was effective across both academic and argumentative writing tasks. Intermediate learners demonstrated stronger effects than basic or advanced learners. In addition, effect sizes did not differ substantially between ChatGPT-based and non-ChatGPT-based systems. Overall, AI-based writing feedback can effectively enhance L2 writing performance across diverse instructional contexts. The findings highlight the importance of pedagogically informed feedback design and feedback orchestration in maximizing the effectiveness of AI-based writing feedback in L2 instruction.

AI-based writing feedback, L2 writing, meta-analysis, moderator analysis, ChatGPT