I. INTRODUCTION
Developing speaking proficiency remains one of the most persistent challenges in English as a Foreign Language (EFL) contexts, where learners have limited opportunities for sustained and meaningful interaction in English. Recent advances in generative AI tools such as ChatGPT have attracted considerable attention for their potential to support interactive language practice and provide immediate feedback (Kohnke et al., 2023; Lim et al., 2023), particularly in EFL contexts where spontaneous, authentic communication opportunities are limited, making speaking one of the most challenging language skills to develop (Kim, 2013; Thornbury, 2005). AI technologies present new possibilities for extending learners language use beyond the traditional educational setting, and AI tools such as ChatGPT, powered by large language models, are capable of generating human-like language (Lim et al., 2023).
Despite these promising affordances, the pedagogical applications of AI tools require careful consideration, particularly in relation to speaking instruction. While ChatGPT has been widely explored as a conversational partner and a source of linguistic support, much of the existing research has focused on general interaction rather than on how such tools can be integrated into structured pedagogical tasks that promote extended oral production (Cheon, 2023; Noh, 2024; Song & Kim, 2025). As a result, there remains a need to examine how AI can support speaking activities that involve idea development, organization, and sustained discourse, which are considered key components of communicative competence.
To support the development of speaking proficiency among EFL learners, mobile-assisted language learning (MALL) and blended learning have also been extensively adopted to expand learning beyond the classroom. MALL emphasizes the use of mobile devices to support ubiquitous and personalized learning experiences (Kukulska-Hulme, 2009; Liu, 2009), along with increased opportunities for practice, interaction, and exposure to the target language outside scheduled class time (Stockwell, 2010). Blended learning combines face-to-face instruction with technology-mediated activities to create more flexible learning environments (Graham, 2006). Building on these approaches, mobile-assisted blended learning (MABL) integrates mobile technologies within structured instructional designs, allowing for meaningful learning activities to continue through in-class and out-of-class environments (Baek & Lee, 2018; Pyo & Lee, 2022). However, prior research suggests that the effectiveness of such environments depends largely on pedagogical integration and task design rather than the technology itself (Burston, 2014). This highlights the importance of carefully structured speaking activities that meaningfully incorporate technological tools.
Storytelling has long been regarded as an effective pedagogical technique for promoting speaking development, as it requires learners to organize ideas, produce extended discourse, and engage in meaningful communication (Thornbury, 2005). In addition to supporting linguistic development, storytelling can enhance learner engagement and reduce speaking anxiety by providing meaningful and contextualized communication tasks. However, storytelling tasks can also pose challenges for EFL learners, particularly in generating ideas, maintaining coherence, and ensuring linguistic accuracy (Haven, 2007; Janeth & Romero, 2023; Maureen et al., 2022; Spencer & Pierce, 2022). In this regard, AI tools such as ChatGPT may serve as a form of scaffolding, supporting both the conceptual and linguistic aspects of speaking (Kohnke et al., 2023).
The above three elements may function complementarily within speaking instruction, from a pedagogical perspective. Storytelling tasks require learners to sustain oral production, organize ideas coherently, and engage in meaningful communication. MABL environments can extend these activities beyond classroom boundaries through continuous interaction and rehearsal opportunities, while ChatGPT may function as a form of scaffolding by supporting idea generation, language refinement, and interactional practice. Together, these components could create conditions that support repeated oral production, learner engagement, and collaborative meaning-making within speaking-focused instruction.
Despite this potential, however, few studies have examined how AI tools such as ChatGPT can be systematically integrated into structured speaking tasks within MABL environments. In particular, there is a lack of research examining how AI-supported storytelling activities can be implemented within a structured blended learning environment to facilitate both speaking proficiency and learner perceptions. Addressing this gap is important not only for understanding the pedagogical potential of AI tools, but also for informing the design of effective speaking instruction in AI-integrated learning environments. Accordingly, this study examines the implementation of ChatGPT-supported storytelling activities within a MABL environment in a university EFL context, focusing on both speaking development and learner perceptions. By examining speaking proficiency and learner perceptions, this study aims to contribute a pedagogically grounded model for the integration of generative AI into speaking instruction.
II. LITERATURE REVIEW
1. Artificial Intelligence in Language Education
Recent advances in AI, especially in large language models (LLMs), are reshaping language learning by enabling more adaptive and interactive experiences. LLMs such as ChatGPT have been widely recognized for their ability to provide sophisticated, immediate feedback and simulate human-like interaction, thereby supporting language learning processes (Zawacki-Richter et al., 2019). As a result, ChatGPT has been frequently used for writing support and grammar correction (Barrot, 2023). However, its ability to simulate naturalistic conversation and offer scaffolded responses implies a strong potential for supporting interactive activities that require sustained discourse (Godwin-Jones, 2022). This potential positions ChatGPT as a particularly valuable tool for speaking practice, especially in contexts where opportunities for authentic communication are constrained.
Recent studies have further emphasized the role of AI tools in supporting learner autonomy and extending language practice beyond classroom environments. By enabling learners to engage with language input and receive immediate feedback, ChatGPT can facilitate self-paced learning and individualized practice, which are often difficult to provide in traditional classroom settings (Kohnke et al., 2023). In particular, the conversational capabilities of ChatGPT allow learners to initiate, sustain, and revisit interactions based on their needs, increasing both the quantity and quality of language use as well as their willingness to communicate (Ahn et al., 2024; Kohnke et al., 2023; Wan & Moorhouse, 2024; Zhang, 2026). However, the effectiveness of AI tools such as ChatGPT is not inherent but depends largely on how they are pedagogically implemented. Concerns have been raised regarding the reliability of AI-generated responses, the risk of superficial interaction, limited personalization, and learners' over-reliance on AI without sufficient reflection (Zawacki-Richter et al., 2019).
These findings suggest that the pedagogical value of ChatGPT lies not in the technology itself, but in how it is embedded within structured instructional tasks that promote sustained and meaningful language use. This is particularly important for speaking instruction, where learners must go beyond brief interaction and develop ideas, organize discourse, and maintain oral production over time.
2. Mobile-Assisted Blended Learning
Mobile-assisted blended learning (MABL) refers to instructional practices that integrate mobile technologies within blended learning environments. In language education, MABL has been used to describe pedagogical arrangements in which classroom instruction is extended through mobile-supported activities, allowing learners to transition between face-to-face interaction and out-of-class practice (Baek & Lee, 2018; Jin, 2014). Prior research emphasizes that the effectiveness of such environments depends not on the presence of mobile technologies alone, but on how instructional activities are coordinated across learning contexts (Baek, 2022). Instead, instructional design, task sequencing, and pedagogical alignment play critical roles in determining the effectiveness of mobile-supported learning (Burston, 2014; Sung et al., 2016). Without careful integration, the use of mobile devices may lead to fragmented learning experiences, superficial engagement, or increased distractions (Stockwell & Hubbard, 2013). These concerns are similarly relevant in mobile-assisted blended learning contexts, underscoring the importance of structured and pedagogically aligned task design.
Empirical studies in EFL contexts have generally reported positive learning outcomes in MABL environments when instructional design is carefully structured. Improvements in speaking fluency and accuracy, as well as increased learner engagement, have been attributed to expanded opportunities for autonomous and sustained practice (Baek & Lee, 2021; Kim, 2014; Pyo & Lee, 2022). These findings reinforce the importance of instructional design and task structure in shaping learning outcomes. Building on this line of research, the present study adopts MABL as an instructional context for implementing ChatGPT-supported storytelling activities. By examining changes in speaking performance and exploring learners' perceptions of these activities, this study investigates how AI-supported storytelling can be systematically integrated into blended EFL instruction.
3. Storytelling for Speaking Development
Storytelling as a pedagogical activity has been examined across a range of language learning contexts, focusing on its potential to support learners linguistic, cognitive, and affective development. It has often been used as a pedagogical task that encourages learners to construct meaning through narrative organization, allowing them to integrate various language skills in a coherent manner (Isbell et al., 2004; Thornbury, 2005). In addition to supporting linguistic development, storytelling has also been examined with regard to learner engagement and affective factors (Maureen et al., 2022). Haven (2007) reported that storytelling activities can foster learner engagement by allowing students to draw on personal experiences or imaginative content, which may contribute to improved motivation and reduced anxiety during oral performance. Storytelling has also been reported to encourage interaction and shared meaning-making as learners continuously build upon each other's narratives in collaborative learning environments (Murphy, 2010).
With the integration of technology, digital storytelling has been explored as a means of supporting narrative construction and language development. Studies have shown that digital storytelling can enhance planning, revision, and reflection processes, while also promoting repeated practice that supports oral development (Janeth & Romero, 2023; Robin, 2008). However, some researchers caution that students may still experience difficulties in organizing narratives, maintaining fluency, and overcoming performance anxiety, and may impose excessive cognitive demands without sufficient scaffolding (Yang & Wu, 2012). These findings highlight the importance of structured task design and sufficient instructional support when implementing storytelling activities in EFL contexts.
Taken together, previous research highlights the pedagogical potential of AI tools, MABL environments, and storytelling-based tasks for supporting speaking development. However, these strands of research have largely been examined in isolation, with limited attention to how they can be pedagogically aligned within a unified instructional framework. More specifically, there is a lack of empirical research on how AI-supported storytelling tasks can be systematically integrated within MABL environments to support both speaking proficiency and learner perceptions. Therefore, to address this gap, the present study investigates EFL university students' perceptions of ChatGPT-supported storytelling in a MABL environment and its effects on their speaking proficiency. The two research questions of this study are as follows:
III. METHOD
1. Participants
The participants of this study were 32 undergraduate students enrolled in a Liberal Arts English course at a university in Goyang, South Korea. All participants were non-native speakers of English and were taking the course as part of their mandatory curriculum. Most participants were sophomores, with two freshmen and one junior. Their ages ranged from approximately 20 to 22 years old. They represented a range of academic departments, reflecting a diverse student population rather than a single disciplinary background, as shown in Table 1.
Despite their differences in academic majors, the participants shared a relatively similar level of English proficiency, which was classified as lower-intermediate, based on institutional placement procedures and the course level associated with the textbook World Link 2 (Hughes et al., 2021). This relative homogeneity in proficiency allowed the study to focus on students' perceptions of ChatGPT-supported storytelling and its effects on speaking proficiency improvement, rather than differences attributable to the participants' various academic backgrounds. Participation in the study was voluntary, and informed consent was obtained from all participants.
Informal conversations at the beginning of the semester suggested that many students were familiar with generative AI tools such as ChatGPT through personal or recreational use. However, most had limited experience using such tools within structured educational or language learning contexts.
2. Data Collection Instruments
1) Materials
The primary tool used in the intervention was ChatGPT, which served three main functions. First, it was used as a speaking partner within mobile-based activities, allowing students to engage in conversational exchanges, rehearse story ideas, and practice responding to follow-up questions related to their stories. Second, it functioned as an idea generator, helping students generate, select, and refine topics and ideas for their storytelling tasks. The representative examples of the ChatGPT are shown in Figure 1.
These activities were based on textbook-related themes, such as personal experiences, introductions of public figures, future technologies, mysteries, hypothetical scenarios, and other related themes. Most storytelling tasks involved introducing and explaining selected topics or materials while incorporating narrative elements and personal interpretation. Third, it supported story construction and revision by providing assistance with vocabulary selection, idea organization, grammatical refinement, and narrative coherence. Students were required to critically evaluate the information provided by ChatGPT and make independent decisions during task completion, both in and outside the classroom. ChatGPT was accessed through the students' mobile devices.
The secondary tool used in the study was KakaoTalk, a widely used messaging application in South Korea. KakaoTalk provided students with an environment where they could collate information, compare their interactions with ChatGPT, and collaborate with their peers both inside and outside the classroom. In addition, KakaoTalk's voice memo function enabled students to engage in asynchronous oral communication by allowing them to record, share, review, and respond to spoken messages at different times rather than through simultaneous real-time interaction. Through these activities, students shared opinions, discussed learning materials and ChatGPT-generated content, exchanged feedback, and continued interaction outside the classroom. The asynchronous nature of these interactions also allowed students greater flexibility in participation and response timing, supporting more continuous interaction within the mobile learning environment (Stockwell & Hubbard, 2013). KakaoTalk was also used as a platform for delivering individualized instructor feedback to the students, and to address student inquiries. Like ChatGPT, KakaoTalk was accessed via the students' mobile devices. Representative examples of the KakaoTalk-supported activities used during the intervention are presented in Figure 2.
The course textbook used in this study was World Link 2 (Hughes et al., 2021), whose thematic units were closely aligned with the storytelling tasks, providing students with familiar and meaningful contexts for oral production. Each unit in the book was split into two separate lessons (A and B), with one lesson covered per week. This two weeks per theme structure allowed students sufficient time for topic familiarization, idea development, and story preparation.
The combined use of ChatGPT, KakaoTalk, the course textbook, and the students' mobile devices allowed the implementation of a MABL environment. Classroom interactions included face-to-face pair and group discussions in which students exchanged opinions, discussed learning materials, and engaged in oral communication activities related to the instructional tasks. These interactions were complemented by ChatGPT's language support provision and KakaoTalk's collaborative capabilities, while KakaoTalk's communicative functions enabled flexible and asynchronous engagement outside the classroom.
2) Questionnaire
To investigate the participants' perceptions of the ChatGPT-supported storytelling activities implemented within a MABL environment, a questionnaire was administered at the end of the course. The questionnaire was designed to capture the students' overall learning experience across four key aspects: 1) interest, 2) perceived difficulty, 3) perceived usefulness, and 4) overall satisfaction. These constructs were adapted from previous studies examining learner perceptions in MABL environments (e.g., Baek & Lee, 2018; Pyo & Lee, 2022), as interest, perceived difficulty, perceived usefulness, and overall satisfaction have frequently been used to examine learners' affective and cognitive responses to technology-supported instructional interventions. Specifically, the questionnaire examined these four aspects in relation to key components of the intervention, including the use of ChatGPT as a learning support tool, the MABL instructional approach, and the storytelling activities. The questionnaire was originally administered in Korean to ensure participant comprehension, and the full translated version is presented in the Appendix.
The questionnaire consisted of both Likert scale items and open-ended questions. The Likert scale employed a four-point format without a neutral midpoint to encourage participants to express clear opinions. The open-ended questions allowed students to elaborate on their perceived experiences, which provided valuable insights into perceived benefits, challenges, and potential limitations. The internal consistency of the questionnaire was assessed using Cronbach's alpha as a general exploratory indicator of reliability (Cronbach's α =.75), based on 31 valid responses across 12 items, with one response excluded by SPSS reliability analysis due to missing data. Since the questionnaire was designed to capture learners' affective responses across three instructional components (ChatGPT, MABL, and storytelling) rather than a single unified construct, the alpha value should be understood as a preliminary estimate of overall internal consistency rather than evidence of a unidimensional scale. Content validity was ensured through alignment of the questionnaire items with the instructional components of the study and relevant literature.
3) Speaking Tests
To examine any changes in students' speaking proficiency, pre- and post-tests were conducted. Both tests consisted of topic-based oral interviews with the instructor, where the topics were related to the themes used in their textbooks. The speaking tests generally required students to introduce, explain, or discuss a selected topic, after which the instructor asked follow-up questions intended to encourage extended oral responses and spontaneous interaction. Each test lasted five minutes, consisting of three minutes allocated to their prepared speech, and two minutes of question-and-answer session about their chosen topic. Although different topics were discussed during the pre- and post-tests, both assessments followed the same format and involved comparable levels of topic explanation, oral development, and spontaneous response production. The tests were assessed using the university's rubric that was adapted from the IELTS speaking test, evaluating students' vocabulary, pronunciation, grammar range and accuracy, fluency and coherence, and task completion.
The speaking assessment rubric is presented in Table 2, and was used for instructional assessment purposes only, and therefore individual rubric components were not analyzed separately in this study. All speaking tests were rated by the researcher using the same rubric to ensure consistency across assessments. While a single rater was used, the same scoring criteria and procedures were applied consistently across both testing points to help ensure scoring reliability. However, because only a single rater was involved in the assessment process, inter-rater reliability measures could not be established.
3. Procedures
The study was implemented as part of a quasi-experimental, one-group pre- and post-test study, conducted over a 16-week instructional period. As this study was designed as a pilot investigation, the one-group design was adopted to explore the feasibility and preliminary effects of integrating ChatGPT-supported storytelling within a MABL framework prior to conducting larger-scale controlled studies.
The participants consisted of two classes, each comprising 16 students. Both classes followed the same instructional procedures, materials, and assessment schedule. The first three weeks featured an induction phase, familiarizing the students with the course structure, storytelling-based activities, and the use of ChatGPT and KakaoTalk as learning tools. The pre-test was administered after the induction period to establish the participants' initial level of proficiency. Following the pre-test, the instructional intervention was implemented through a two-session cycle for each thematic unit, consisting of Session 1 (Lesson A) and Session 2 (Lesson B).
As shown in Figure 3, session 1 focused on introducing the topic of the unit and supporting initial idea development for storytelling. During the in-class section, the instructor introduced the topic and key language, guided students through activities such as picture description, vocabulary work, multimedia comprehension, and role-play or information-gap tasks. The purpose of these activities was to build background knowledge and to prepare students for subsequent storytelling tasks. General teacher feedback was provided between the activities. As for the out-of-class section, students engaged in mobile-focused tasks. They first individually used ChatGPT's voice function to request information about the unit's theme, and to decide on a topic for their story composition. They then shared their interaction with ChatGPT with their group members, utilizing KakaoTalk's group chat function, before discussing which topic to pursue in the second session through the voice memo feature. The instructor monitored this process and provided focused, individualized feedback primarily through text-based messages on KakaoTalk, addressing topic development, appropriate story length, vocabulary use, grammatical issues, and overall clarity of expression when necessary.
Session 2 emphasized story composition, presentation, and revision. The in-class activities featured brainstorming, story composition, and storytelling activities, based on the discussions from Session 1. The entire story composition process was recorded through KakaoTalk as participants uploaded information generated through ChatGPT throughout the composition process, which was monitored by the instructor. During this process, students also exchanged opinions and suggestions with peers through face-to-face group interaction and discussion. The story drafts were checked, and instructor feedback was provided to each group before finalization, focusing on story organization, appropriate length, vocabulary refinement, grammatical accuracy, and coherence of presentation. Once all stories were approved, students were reorganized into storytelling groups, each consisting of members from different composition groups. The transition from story composition groups to storytelling groups was systematically structured to ensure that each storytelling group consisted of members who worked on different stories. The students were also to ask at least one question to their favorite storyteller, which encouraged active listening, and increased opportunities for oral interaction.
For the out-of-class activities of Session 2, students focused on reflection and improvement, again using KakaoTalk's communication capabilities. The first main activity was to provide a brief verbal summary of the stories, before discussing their favorite stories, providing reasons for their preferences. The final activity consisted of individually improving a story of their choice through ChatGPT's assistance and sharing their results. Individualized instructor feedback was again provided primarily through text-based KakaoTalk messages.
Figure 4 shows a more detailed outline of the instructional sequence across the two sessions. Each session followed a structured progression consisting of warm-up, main, and wrap-up activities, with clear distinctions between in-class and out-of-class learning environments. In both sessions, peer feedback and teacher feedback were deliberately embedded at multiple stages of the instructional process. The above figure also highlights how speaking practice was distributed across different activity types, including comprehension, collaborative discussion, storytelling, and story revision. This instructional treatment was designed to provide sustained speaking opportunities throughout the sessions, from the fourth week onwards.
The post-test was conducted during the 16th week, using the same format and procedures as the pre-test. The post-intervention questionnaire was distributed during the same week, to collect data on students perceptions regarding the ChatGPT-supported storytelling activities and the MABL environment.
4. Methods of Data Analysis
The quantitative data collected from the questionnaire, along with the pre- and post-tests, were analyzed using SPSS. More specifically, frequency analyses were conducted for the four affective factors: interest, perceived difficulty, perceived usefulness, and overall satisfaction, to examine the participants' perceptions of the ChatGPT-supported MABL environment, and the implementation of storytelling activities. Considering the exploratory nature and relatively small sample size of this pilot study, descriptive statistics were primarily used to identify general trends in learner perceptions rather than establishing statistically generalizable differences among the three components. The responses to open-ended questions in the questionnaire were used to explain and contextualize quantitative findings, providing deeper insight into the students' perceptions. These responses were reviewed descriptively to identify recurring observations and commonly reported experiences related to the instructional activities. Representative responses were used to support and contextualize the quantitative findings. The pre- and post-tests were analyzed through paired-samples t-tests, to investigate whether there was a statistically significant improvement in the students' speaking proficiency.
IV. RESULTS
1. Students' Perceptions
To address the first research question, frequency analyses were conducted on the students' responses across the four affective factors: Interest, Perceived Difficulty, Perceived Usefulness, and Overall Satisfaction. Students' responses were analyzed separately for the three instructional components: ChatGPT, MABL, and storytelling activities. Although ChatGPT, MABL, and storytelling were implemented as an integrated framework, learner perceptions were examined separately for each component because each element represents a feature that distinguishes the present intervention from more conventional EFL speaking instruction, where such components are not typically integrated. Examining the components individually therefore allows a clearer understanding of how learners responded to different aspects of the intervention, which may provide granular pedagogical insights for future research and instructional design. The separate treatment of the components should be understood not as a departure from the integrated design, but as a way of examining how learners perceived the distinct functions within the broader instructional framework.
Table 3 presents the frequency distribution of the participants' interest for ChatGPT, MABL, and storytelling activities. The results show that a large majority of students responded positively for ChatGPT (93.8%) and MABL (90.7%), with only a small proportion indicating negative responses. In addition, a higher proportion of students selected "Very Interesting" for MABL compared to ChatGPT, suggesting slightly stronger engagement with the MABL environment. This pattern indicates that both ChatGPT and MABL were generally well received by the participants, which is consistent with previous research suggesting that AI-supported interaction and mobile-based learning environments can enhance learner engagement when appropriately integrated into instructional tasks (Ahn et al., 2024; Baek & Lee, 2021; Pyo & Lee, 2022).
In contrast, storytelling activities, while still being positively received by most, showed relatively more negative responses (25%). This suggests that while storytelling was generally engaging, it may not have appealed equally to all learners. This may be attributed to some students finding the topics provided uninteresting or the cognitive demands of the task, as a few have mentioned in their comments.
Table 4 summarizes the frequency distribution for Perceived Difficulty across the three instructional aspects. The results show that the participants generally did not perceive the activities as overly difficult. For ChatGPT, 93.8% of the students selected "Easy" or "Very Easy", and no participants reported it as "Very Difficult," indicating that interacting with ChatGPT was not perceived as burdensome.
For MABL, perceived difficulty was slightly more varied, with 18.7% of the students claiming that it was "Difficult," suggesting that some challenges were encountered in the MABL environment. The open-ended question responses further indicated that these difficulties were often related to scheduling issues, particularly in coordinating participation among group members. For example, one participant noted that "it was sometimes difficult to coordinate discussion times with group members outside class." This finding is consistent with previous research identifying time coordination as a common challenge in MALL and MABL environments (Kukulska-Hulme, 2009; Stockwell, 2010).
Storytelling was perceived as the most challenging component, with 28.1% of the students reporting negative responses. However, the majority (71.9%) still evaluated the activity positively. This may reflect that while storytelling demanded greater effort, it remained manageable for most learners. Some participants reported difficulty in understanding peers due to the use of varied vocabulary generated through ChatGPT, while others noted challenges in coordinating ideas during the story composition process. For instance, one participant commented that "some expressions generated through ChatGPT were unfamiliar, which occasionally made communication difficult," while another reported that it was "hard to combine and organize many ideas."
The frequency analysis results for Perceived Usefulness are presented in Table 5. The results show generally positive responses across all components, with 93.8% of participants reporting positive evaluations for ChatGPT, MABL, and storytelling activities.
Notably, despite being perceived as the least interesting and most difficult component, storytelling received equally high usefulness ratings. This may indicate that learners recognized the value of storytelling activities even when they were cognitively demanding. This is also reflected in the open-ended responses, with one participant stating that "the storytelling activities were difficult at first, but they helped me organize my ideas and speak more naturally." This pattern suggests that task difficulty did not necessarily reduce learners' perceived usefulness and may instead have contributed to their perception of learning value.
One possible explanation for this pattern is that storytelling required learners to engage more actively in idea organization, extended oral production, and collaborative interaction than the other instructional components. Although these demands may have increased the perceived difficulty, they may also have contributed to stronger perceptions of usefulness and satisfaction when learners were able to successfully complete the tasks with sufficient support. ChatGPT may also have functioned as a form of pedagogical scaffolding by supporting idea generation, rehearsal, and language refinement throughout the storytelling process, while the MABL environment extended opportunities for interaction and practice beyond the classroom. Together, these elements may have helped learners engage more meaningfully with cognitively demanding speaking activities. This interpretation aligns with previous research suggesting that cognitively demanding speaking tasks can promote deeper processing and more meaningful learning experiences when supported by adequate scaffolding and structured instructional design (Thornbury, 2005; Yang & Wu, 2012).
Table 6 presents the frequency distribution of overall satisfaction across the three instructional components. The results show consistently high levels of satisfaction, with ChatGPT receiving the highest positive responses (96.9%), followed by MABL (93.8%). Storytelling also received strong evaluations, with the highest proportion of "Very Satisfied" responses (53.2%), indicating a particularly high level of satisfaction with the activity.
This overall pattern may reflect that students responded positively to the integration of ChatGPT, MABL, and storytelling activities within the instructional design. Notably, despite being perceived as more challenging in earlier results, storytelling demonstrated strong satisfaction outcomes, indicating that learners valued the activity even when it required greater effort. This may suggest that cognitively demanding tasks can be associated with higher satisfaction when they are supported by structured guidance and meaningful interaction. This finding is consistent with previous research on integrated instructional approaches combining technology, task-based learning, and mobile support (Baek & Lee, 2018).
Across the four affective dimensions, several consistent patterns emerge. First, positive responses were dominant across all dimensions, indicating generally favorable learner perceptions. Second, storytelling, while perceived as more challenging than ChatGPT or MABL, received the strongest ratings for perceived usefulness and overall satisfaction, suggesting that learners recognized its pedagogical value. Third, ChatGPT was evaluated positively across all dimensions, reinforcing its role as a stable and supportive learning tool within the instructional framework.
Taken together, these findings suggest that the integration of ChatGPT within a MABL environment may have helped mitigate, although not entirely eliminate, some of the challenges associated with storytelling tasks, such as cognitive load and performance anxiety. By supporting idea generation, rehearsal, and extended practice beyond the classroom, ChatGPT may have contributed to conditions that enabled learners to engage more effectively with storytelling as a speaking activity (Haven, 2007; Wan & Moorhouse, 2024).
2. Effects on Speaking Proficiency
To address the second research question, a paired-samples t-test was conducted to examine whether participants' speaking proficiency improved. As shown in Table 7, the results indicated a statistically significant difference between the pre-test and post-test speaking scores, t(31)= -4.08, p <.001. The participants' post-test scores (M = 62.53, SD =10.19) were significantly higher than their pre-test scores (M = 52.11, SD =12.90), indicating an overall improvement in speaking proficiency.
The magnitude of the difference was medium to large, as indicated by Cohen's d = 0.72, suggesting a practically meaningful effect size. These results align with previous research indicating that storytelling-based speaking tasks can promote speaking development by encouraging extended oral production, narrative organization, and repeated practice (Isbell et al., 2004; Thornbury, 2005). Additionally, the structured integration of MABL and storytelling activities may have contributed to the observed improvement by enabling learners to engage in planning, rehearsal, and revision across both in-class and out-of-class learning environments (Baek & Lee, 2018; Pyo & Lee, 2022).
One possible explanation for the observed improvement is that the intervention provided students with sustained opportunities for extended oral production across both classroom and mobile-supported environments. Unlike more conventional speaking activities that are often limited to brief classroom interaction, the present instructional framework encouraged students to engage in idea generation, rehearsal, discussion, storytelling, and revision across multiple stages of the learning process. Furthermore, the MABL environment extended opportunities for speaking practice beyond classroom boundaries, allowing students to continue interaction and collaborative discussion asynchronously through mobile-supported activities.
ChatGPT may also have functioned as a form of pedagogical support within these activities by assisting learners with idea development, conversational rehearsal, and language refinement. Rather than replacing classroom interaction, ChatGPT was embedded within structured speaking-oriented tasks that required learners to actively organize ideas, negotiate meaning, and participate in collaborative storytelling activities. These findings are consistent with previous research suggesting that AI-supported speaking activities may be most effective when integrated into structured instructional frameworks that promote sustained learner engagement and repeated speaking practice.
These findings are also consistent with emerging research on AI-supported oral interaction. Previous studies have reported that tools such as ChatGPT can facilitate speaking development by providing accessible opportunities for interaction and reducing pressure associated with face-to-face communication (Godwin-Jones, 2022; Zhang, 2026). Prior research further suggests that such AI-supported speaking activities may be particularly effective when embedded within structured instructional designs that promote learner engagement and repeated speaking practice (Ahn et al., 2024; Kohnke et al., 2023). Overall, these findings suggest that the observed improvement in speaking proficiency may have been supported by the integration of ChatGPT-supported storytelling activities within a MABL environment, although the absence of a control group limits causal interpretation.
V. CONCLUSION
This study reports on a pilot investigation of the implementation of ChatGPT, MABL, and storytelling activities in a university EFL context, focusing on students affective responses and speaking proficiency improvement. The findings indicate that students demonstrated generally positive perceptions across all affective dimensions and showed a statistically significant improvement in speaking proficiency, although the absence of a comparison group limits causal interpretation. These findings give rise to three key implications.
First, generative AI tools such as ChatGPT may be most pedagogically effective when embedded within structured speaking-oriented tasks rather than used as standalone conversational tools. Carefully designed AI-supported activities may help promote more active learner participation during speaking tasks. Second, the study highlights the pedagogical value of integrating AI-supported activities within a MABL framework. The findings suggest that connecting classroom-based interaction with mobile-supported practice may provide more continuous opportunities for speaking engagement beyond the classroom. Third, the findings highlight the potential role of storytelling as a speaking task within AI-enhanced speaking instruction. Although storytelling was perceived as relatively challenging, learners also evaluated it positively in terms of usefulness and satisfaction, suggesting that structured support and collaborative activities may help learners engage more actively with cognitively demanding speaking tasks.
Together, the findings of this study indicate that integrating ChatGPT, MABL, and storytelling activities may provide a promising, pedagogically grounded approach to EFL speaking instruction, provided that tasks are carefully designed. More broadly, the results highlight that the effectiveness of AI tools in language learning appears to depend not on the technology itself, but on how it is integrated within structured instructional designs that promote interaction, support, and learner participation. These implications may extend to future developments in AI-enhanced language learning, particularly in relation to how emerging technologies can be integrated within structured instructional frameworks. As AI tools continue to evolve, the importance of pedagogically grounded design will remain central to developing effective speaking-oriented learning environments.
Despite these contributions, several limitations of this study should be acknowledged. The use of a one-group pre- and post-test design, even as a pilot, limits the ability to attribute improvements solely to the intervention. In other words, because repeated speaking practice was embedded throughout the instructional cycle, it is difficult to distinguish the effects of the intervention itself from broader practice-related improvement in the absence of a comparison group. Additionally, the relatively small sample size from a single institutional context affects the generalizability of the findings, and the relatively small number of items in the questionnaire, combined with the use of a four-point Likert scale, and the lack of a formal qualitative coding procedure for the open-ended responses, may have been insufficient to capture more nuanced aspects of learner perceptions and experiences. Also, although KakaoTalk interactions formed an important part of the instructional process, these interaction data were not systematically collected or analyzed qualitatively. Furthermore, the use of a single rater for speaking assessment may introduce potential bias, particularly in the absence of inter-rater reliability measures, even though consistent scoring procedures were applied throughout the study.
Future research should consider larger-scale studies incorporating control groups to improve generalizability and isolate the effects of individual components of the intervention. Further investigation into optimal task design, particularly in balancing cognitive demand and instructional support, would also be valuable. In addition, incorporating qualitative analyses of learner interactions and feedback exchanges in the mobile-supported learning environment may provide deeper insights into peer scaffolding, collaborative processes, and patterns of AI-supported engagement. Finally, exploring a wider range of AI tools may help identify more effective approaches to supporting oral language development.
PDF Links
PubReader
ePub Link
Full text via DOI
Download Citation
Print
