The rapid development of generative artificial intelligence has created new opportunities for innovation in teacher education, dramatically changing how pre-service teachers acquire new skills and knowledge. In the information age, pre-service teachers' information technology-enhanced instructional design competence has received widespread attention. This study introduced the generative artificial intelligence represented by ChatGPT to innovate the traditional teaching mode of pre-service education. Learners use ChatGPT to automatically generate courseware and revise courseware according to the framework provided by concept maps. To promote the depth of thinking and motivation of pre-service teachers in the revision process, we proposed an integrated approach with the peer assessment and ChatGPT. The researchers conducted a quasi-experimental study at a university to investigate the effectiveness of this learning strategy. One class of pre-service teachers (N = 46) was the experimental group using peer assessment-based concept mapping-supported ChatGPT generated content (PA-CGPT), and the other class (N = 45) was the control group using conventional concept mapping-supported ChatGPT generated content (C-CGPT). The study suggests that students in the experimental group exhibited significant improvements in practical skills, learning motivation, higher-order thinking tendencies, satisfaction, and reduced cognitive load. Furthermore, in-depth interviews were conducted with both groups of students to examine their learning outcomes and discuss implications for future research.

'Emerging Technologies in Second Language Teaching and Learning' examines how artificial intelligence and other digital innovations are remaking language education across K–12 and higher education contexts. As technologies such as generative AI, adaptive learning systems, virtual reality, and digital storytelling become increasingly integrated into classrooms, educators face new opportunities—and challenges—in designing meaningful language-learning experiences. This edited volume brings together international scholars and practitioners to explore how emerging technologies can support multilingual learners through personalized learning, immersive environments, and innovative pedagogical approaches. Grounded in established theories of second language acquisition, applied linguistics, and instructional design, the chapters combine research-based analysis with real-world classroom applications. Contributors examine topics including AI chatbots for language practice, AI-generated instructional materials, adaptive learning platforms, culturally adaptive digital storytelling, virtual reality–based language tasks, and open educational resources that expand access to language learning. Aside from technological innovation, the volume critically addresses the pedagogical and ethical issues of AI-enhanced education. Chapters explore issues such as algorithmic bias, academic integrity, digital equity, and teacher agency, as well as the importance of maintaining human-centered learning environments in increasingly automated educational systems. Across a variety of contexts—from secondary classrooms to university language programs—the contributors highlight how instructors can thoughtfully integrate technology to enhance interaction, creativity, and communicative competence. Integrating theory, research, and practice, this book offers a timely resource for language educators, instructional designers, teacher educators, and researchers in TESOL, applied linguistics, multilingual education, and educational technology. It also serves as a valuable guide for graduate students and academic leaders seeking responsible, evidence-informed approaches to innovation in language teaching.

The development of education is contemporary.Human-machine collaborative teaching,as an important practice of intelligent education,carries the unique educational significance and value in the intelligent era,which has been widely explored.In order to promote the theoretical exploration of man-machine collaborative teaching in the increasingly normalized practice,this paper firstly discussed the educational value carried by human-machine collaborative teaching in implementing national strategies,innovating educational practices,and promoting teacher development,and thereby systematically sorted out the current situation of practice.Then,based on the ideological enlightenment of the synergy theory,the human-machine collaborative teaching was deconstructed as"teachers and intelligent machines cooperating to jointly promote students'learning and achieve teaching goals",thereby analyzing the practical requirements of human-machine collaborative teaching as"adhering to one foundation,maintaining one center,and closely focusing on one key point".Finally,based on practical requirements,the study put forward specific suggestions to promote teachers to correctly position their roles in human-machine teaching,standardize the exploration of human-machine collaborative teaching patterns,and actively carry out research-based practices.

This study investigates the potential of Generative Student Agents (GenSAs), implemented via the MetaClass system, to alleviate teaching anxiety among pre-service teachers (PTs) in a university-based teacher education program. Grounded in situated learning and productive unsettlement, we conceptualize anxiety not only as a barrier but also as a driver of professional growth. Using a one-group pretest-posttest quasi-experimental design with 23 PTs, results show a significant reduction in overall teaching anxiety alongside nuanced shifts across specific dimensions. Participants also reported positive attitudes toward GenSA integration, highlighting its pedagogical value. These findings provide empirical evidence that GenAI-driven simulations can both mitigate and productively reshape teacher anxiety. Implications for AIED, particularly in designing affect-aware GenAI learning environments, and future research directions are discussed.

This special issue explores the theories, design, and applications of multiple intelligent agents (MIA) in education. Ten selected papers examine how multi-agent systems (MAS) enable specialized agents to collaborate in ways that surpass single-agent capabilities. Contributions span systematic reviews, controlled experiments, and system evaluations, covering themes such as design frameworks, assessment and feedback, cognitive and affective dimensions, language learning, and human-social factors. Key findings highlight the importance of task specialization, the integration of established learning theories, and the critical role of human-centred and ethical design. Together, these contributions demonstrate the potential of MAS to create adaptive, context-sensitive learning environments while identifying future directions related to transparency, fairness, and reliability in increasingly complex agent coordination.

Personalized learning aims to provide adaptive support that responds to individual differences during learning. While existing systems have made progress in content recommendation and task sequencing, they pay limited attention to how learners regulate their learning through strategies during interaction, which constrains support for deeper learning and self-regulation. This study examines the feasibility of modeling learning strategies using a human-centered reinforcement learning framework. It conceptualizes learning as an interactive process in which observable learning states inform strategy choices and learning outcomes provide feedback. Using large-scale learning trace data, the study maps learning states, strategies, and learning gains into an interpretable state-strategy-reward structure and evaluate the framework across multiple reinforcement learning algorithms. The results show that the framework generates differentiated strategy choices across learning states and shows stable reward improvement across algorithms. While model-generated strategies closely align with learners' short-term outcomes, systematic differences emerge in long-term learning trajectories across learning contexts. These findings highlight the potential of reinforcement learning as a simulation tool for analyzing learning strategy dynamics in adaptive learning environments.

Intelligent tutoring systems increasingly rely on large language models (LLMs) to simulate learner behaviors for personalized feedback and adaptive instruction. However, achieving realistic student simulation remains challenging, especially in modeling how learners of different proficiency levels reason, err, and improve through feedback. LLMs have shown strong capabilities in natural language generation but remain limited in simulating the nuanced behavioral simulation trajectories underlying human learning, especially in educational contexts. Existing role-playing approaches often rely on fixed personas and lack authentic behavioral simulation, leading to shallow or inconsistent emulation of learner behavior. In this paper, we present a structured framework to structured framework for data-driven behavioral simulation through LLM-based role-playing agents. Our approach involves three key components: (1) extracting students’ problem-solving trajectories and feedback patterns from real-world tutoring data, (2) modeling domain-specific knowledge dependencies and reasoning strategies, and (3) simulating observed learning behaviors through natural language narratives for fine-tuning LLMs. Empirical results on student datasets demonstrate that our method enables the agent to replicate diverse reasoning levels and characteristic errors with higher fidelity, offering a data-driven pathway toward learner-aware AI tutoring systems, while also suggesting a cognitively inspired direction for future research. All source codes and datasets are publicly available at: https://github.com/qi-github-ui/CogTutor.git.

This study designed an integrated concept mapping and generative artificial intelligence (generative AI) approach to enhance pre‐service teachers' (PSTs) digital storytelling (DST) outcomes and critical thinking. A quasi‐experimental study was conducted with 62 PSTs, where an experimental group ( n = 30) employed the integrated concept mapping and generative AI approach and a control group ( n = 32) used a standalone generative AI approach. Results indicated that the integrated approach produced broad improvements across most DST outcome dimensions, specifically enhancing ‘overall’ quality, ‘accuracy’, ‘completeness’, ‘innovation’ and ‘interaction’. However, the integrated approach did not yield a greater improvement in critical thinking tendencies compared to the control group; in fact, the control group demonstrated a significantly larger gain over time. Qualitative analysis of reflective reports revealed a key difference in PSTs' perceptions: the experimental group prioritised human–machine interaction as the most essential competency for DST creation with AI, whereas the control group emphasised critical thinking. This suggests that the pedagogical design, specifically the use of concept mapping as a structural scaffold, effectively shapes how PSTs perceive and engage with technology, guiding them towards more reflective and efficient learning with generative AI, even if its impact on critical thinking tendencies is more complex. The study offers relevant insights for educational practitioners seeking to utilise generative AI to develop PSTs' specific DST competencies and their understanding of human–AI collaboration.

To address the challenges of high uncertainty, intersubject variability, and inefficiency in multifeature utilization in motor imagery electroencephalogram classification, this study proposes a multiview transfer Takagi-Sugeno-Kang (TSK) fuzzy classifier with soft-variable embedded and discriminative structural preservation (MVT-TSK-SVDS). First, a transfer learning mechanism incorporating soft-variable embedding in the consequent part is developed. This mechanism establishes cross-domain correlations via a shared consequent and representation matrix. Within this framework, soft-variable embedding and low-rank constrained discriminative learning work in concert to effectively capture supervision information and cross-domain relationships. Second, a local-global structural preservation term incorporating graph embedding and low-rank constraint is implemented to maintain local discriminative information from the source domain while integrating global geometric patterns across all data. Third, a multiview adaptive learning framework is designed to address feature representation diversity and information loss during knowledge transfer. MVT-TSK-SVDS dynamically optimizes view-specific contributions through an entropy maximization criterion while ensuring collaborative decision via consistency constraints. The experimental results validate strong generalization between and across datasets. Our model achieves 62.16% and 72.71% accuracy in cross-subject tasks on BCI-IV 2a and OpenBMI, respectively. In cross-dataset evaluations, it attains 62.75% accuracy on BCI-IV 2a to OpenBMI and 65.08% accuracy on OpenBMI to BCI-IV 2a, respectively.

Educational dialogue classification is a critical task for analyzing classroom interactions and fostering effective teaching strategies. However, the scarcity of annotated data and the high cost of manual labeling pose significant challenges, especially in low-resource educational contexts. This article presents the EduDCM framework for the first time, offering an original approach to addressing these challenges. EduDCM innovatively integrates distant supervision with the capabilities of Large Language Models (LLMs) to automate the construction of high-quality educational dialogue classification datasets. EduDCM reduces the noise typically associated with distant supervision by leveraging LLMs for context-aware label generation and incorporating heuristic alignment techniques. To validate the framework, we constructed the EduTalk dataset, encompassing diverse classroom dialogues labeled with pedagogical categories. Extensive experiments on EduTalk and publicly available datasets, combined with expert evaluations, confirm the superior quality of EduDCM-generated datasets. Models trained on EduDCM data achieved a performance comparable to that of manually annotated datasets. Expert evaluations using a 5-point Likert scale show that EduDCM outperforms Template-Based Generation and Few-Shot GPT in terms of annotation accuracy, category coverage, and consistency. These findings emphasize EduDCM’s novelty and its effectiveness in generating high-quality, scalable datasets for low-resource educational NLP tasks, thus reducing manual annotation efforts.

Generative Artificial Intelligence (GenAI) stands as a cornerstone of the technological revolution, significantly impacting the global educational landscape. This prompts worldwide governments and educational institutions to craft strategic frameworks. This study aims to analyze GenAI's influence on the education system, particularly focusing on transformations in educational paradigms, modalities, pedagogical logics, and educational contexts. It seeks to establish a transformation action framework for the education system in the GenAI era. Utilizing Meta-ethnography, the research synthesizes, analyzes and interprets 11 policy and guideline documents from UNESCO, OECD, ministries of education and universities, which reveal trends towards personalized and interactive educational forms, shifts in the role of the teacher, and updates in student learning modes. The study explores GenAI's integration into education at macro, meso, and micro levels. At the macro level, the framework identifies how GenAI drives a productivity revolution and reshapes human resource demands, alongside societal attitudes and educational actions adapting to this transformation. At the meso level, it reflects on educational pattern and logic shifts, delving into the evolution of educational modalities, entities, media and content. At the micro level, it deconstructs new teaching and learning scenarios in the GenAI era, closely examining the evolution of the role of the teacher and student learning modes, scrutinizing the core value of education as a fundamental human right and constructing a vision for future education in the GenAI era. The findings underscore the need for comprehensive transformation in the education system to adapt to GenAI-driven changes, updating educational content and methods to enhance teaching efficiency and quality as well as fostering holistic student development. These insights offer theoretical and practical guidance for the educational sector to respond to GenAI-driven technological changes, aiming to equip the education system to overcome challenges, seize opportunities and prepare talents needed for the future society.

This paper proposes a creative, AI-driven integrated learning evaluation model that assesses metacognition and deeper learning through multimodal data analysis. Existing research faces challenges in learning data effectiveness, limited measuring tools and methods, and absence of feedback optimization loop. To address these issues, our ubiquitous multidimensional model integrate conscious and unconscious learning data using hybrid reasoning neural networks, generating interpretable representations aligned with metacognitive and deeper learning elements. This approach enables comprehensive assessment, automated feedback, and iterative optimization to enhance students' selfregulation and support students' personal learning needs. By advancing AI in Education (AIED), our integrated evaluation model explores new path for dynamic educational interventions and personalized pedagogies. This research contributes to the field by addressing validity issues, integrating qualitative and quantitative methods, and the loop with feedback optimization.

In the field of education, the think-aloud protocol is commonly used to encourage learners to articulate their thoughts during the learning process, providing observers with valuable insights into learners' cognitive processes beyond the final learning outcomes. However, the implementation of think-aloud protocols faces challenges such as task interference and limitations in completeness and authenticity of verbal reports. This study proposes a method called Cognitive Echo, which leverages large language models (LLMs) trained with simulated student experiences to enhance the completeness and authenticity of think-aloud verbalizations. LLMs have been demonstrated to simulate human-like behaviour more effectively by memorizing experiences. In this work, we introduce specific learner roles and train the LLMs to act as distinct learners. Our method involves integrating transaction data from learners' interactions with a tutoring system and the tutor's content to create interactive experiences between learners and teachers, thereby training the model to become simulated students with learning experiences. To investigate the effectiveness of this approach, we designed a test playground based on the retrospective think-aloud protocol and examined how LLM-trained simulated students improve cognitive process transparency and generalization of learning strategies. The study found that Cognitive Echo not only reveals what simulated students genuinely think about their learning experiences but also enables them to transfer their different cognitive strategies to new tasks. By training simulated students on real learning behaviour data to ensure their cognitive processes reflect authentic learner experiences, this approach will extend think-aloud protocols to more practice-oriented applications.Practitioner notes What is already known about this topic Think-aloud protocols are widely used in educational settings to explore students' cognitive processes by asking them to verbalize their thoughts while solving problems, but they are prone to issues like task interference and incomplete data reporting. Existed applications of simulating student cognition in educational research are rigid and less adaptive to individual learner characteristics. Artificial intelligences, especially large language models, have shown promise in educational contexts, particularly for simulating human-like behaviours. What this paper adds This paper introduces the concept of Cognitive Echo, a method that integrates LLM-powered simulated students into think-aloud protocols, which addresses the limitations of traditional verbalization-based methods by leveraging retrospective data. The study shows that LLMs, when fine-tuned with authentic learner experiences, can replicate distinct human-like cognitive processes, enabling a more complete and authentic simulation of how students think and solve problems. It demonstrates how the use of LLMs to simulate students' cognitive processes can enhance the transparency and completeness of think-aloud protocols by allowing researchers to capture cognitive strategies and behaviours that would otherwise go unspoken. Implications for practice and/or policy Teacher training programmes can benefit from integrating LLM-based simulated students, which enable preservice teachers to practice responding to a wide range of cognitive processes and challenges without the constraints of real-time think-aloud tasks. The Cognitive Echo method, by offering a more authentic and less intrusive way of capturing student cognition, can be applied in teacher training scenarios where simulation of real-world classroom dynamics is crucial for developing pedagogical skills. The use of Cognitive Echo could help in the creation of digital twins of educational scenarios, facilitating research into complex educational issues (eg, bullying and learning disabilities) through simulations that model real-world interactions.

The introduction of large language models (LLMs) may change future pedagogical practices. Current research mainly focuses on the use of LLMs to tutor students, while the exploration of LLMs’ potential to assist teachers is limited. Taking high school mathematics as an example, we propose a method that utilizes LLMs to enhance the quality of teaching plans through guiding the LLM to simulate teacher-student interactions, generate teaching reflections, and subsequently direct the LLM to refine the teaching plan by integrating these teaching process and reflections. Human evaluation results show that this method significantly elevates the quality of the original teaching plans generated directly by LLM. The improved teaching plans are comparable to high-quality ones crafted by human teachers across various assessment dimensions and knowledge modules. This approach provides a pre-class rehearsal simulation and ideas for teaching plan refinement, offering practical evidence for the widespread application of LLMs in teaching preparation.

As the demand for precision, efficiency, and low-cost solutions in educational classification tasks continues to grow, enhancing model performance has become a critical focus of research. While large language models excel in these tasks, their high cost and resource requirements limit widespread application. This study proposes a Knowledge-Enhanced Distillation (KED) method, utilizing ChatGPT-4, ChatGPT-4o, and Llama3 as teacher models, and three different sizes of BERT models as student models. The method was validated across three real-world educational datasets. The results demonstrate that the KED method significantly improves the accuracy and Fl scores of small models in educational text classification tasks, while also substantially reducing computational costs and resource consumption. Notably, the KED method shows exceptional performance in scenarios involving few-shot learning and class imbalance. The innovation of this study lies in applying the KED method to educational classification tasks, filling a gap in current research and highlighting its significant potential for practical application in educational contexts.

ABSTRACTBackgroundChatGPT, as a cutting‐edge technology in education, is set to significantly transform the educational landscape, raising concerns about technological ethics and educational equity. Existing studies have not fully explored learners' intentions to adopt artificial intelligence generated content (AIGC) technology, highlighting the need for deeper insights into the factors influencing adoption.ObjectivesThis study aims to investigate higher education learners' adoption intentions towards AIGC technology, with a focus on understanding the underlying reasons and future prospects for its application in education.MethodsThe research is divided into two phases. First, an exploratory analysis involving practical activities and interviews develops an action decision framework for AIGC adoption. Second, a confirmatory analysis using fuzzy‐set qualitative comparative analysis on 233 valid questionnaires identifies six configurations associated with high adoption intentions, emphasising the roles of AI literacy and perceived behavioural control.Results and ConclusionsThe study reveals key factors influencing AIGC adoption, including the importance of AI literacy and perceived behavioural control. It provides actionable insights for educators and learners to prepare for and effectively integrate AIGC technology, ensuring equitable and adaptive educational practices.

Although music education is considered a fundamental right for all, disparities in access remain widespread. Learners often face unequal opportunities shaped by their family backgrounds and prior experiences. This study explored the potential of AI integration in blended learning to promote inclusive and accessible music theory education. By utilizing AI-driven feedback in blended learning (AF-BL), students benefit from tailored learning experiences that promote equal opportunities for growth and reflection. A total of 43 students from a public university in China participated in a 4-week music theory course. They were divided into two groups: an experimental group (N = 22) utilizing the AF-BL method, and a control group (N = 21) following the conventional blended learning (C-BL) method. The results demonstrated that the AF-BL method significantly improved learners' music theory learning outcome and perceptions, compared to the C-BL method. Interviews with participants further highlighted the inclusivity and accessibility of the AF-BL approach, noting its ability to cater to diverse learning needs and provide equal learning opportunities for all students. The findings highlight the potential of AI in creating equitable and inclusive educational experiences, suggesting promising directions for future research and practical applications in music theory education.

Prior research has focused on students' affectional states during learning, recognizing that affect can elevate learning and cognitive development. This study hypothesized and investigated the bidirectionality or reciprocity of the relationships between students' affects and performance in reading comprehension. Mixed methods were used in the study. 41 3rd grade (9-year-old) students went through mind mapping in reading intervention for one semester. Pre- and post-intervention surveys and post-interviews were conducted to investigate their affectional and self-efficacy levels in reading comprehension. Their mind maps were also collected to evaluate their performance in reading comprehension. The study results revealed that the use of mind mapping in reading activities led to a widespread sense of positive affect. Furthermore, positive affect was found to be influenced by reading comprehension, and reading comprehension could contribute to changes in affect by enhancing reading performance and increasing self-efficacy. This study elevates the understanding of the mind mapping in reading strategies, provides references for the affectional design of such activities, and derives design and enactment principles to transform mind mapping in reading activities.

With the continuous advancement of educational technology, the demand for Large Language Models (LLMs) as intelligent educational agents in providing personalized learning experiences is rapidly increasing. This study aims to explore how to optimize the design and collaboration of a multi-agent system tailored for Socratic teaching through the integration of LLMs and knowledge graphs in a chain-of-thought dialogue approach, thereby enhancing the accuracy and reliability of educational applications. By incorporating knowledge graphs, this research has bolstered the capability of LLMs to handle specific educational content, ensuring the accuracy and relevance of the information provided. Concurrently, we have focused on developing an effective multi-agent collaboration mechanism to facilitate efficient information exchange and chain dialogues among intelligent agents, significantly improving the quality of educational interaction and learning outcomes. In empirical research within the domain of mathematics education, this framework has demonstrated notable advantages in enhancing the accuracy and credibility of educational interactions. This study not only showcases the potential application of LLMs and knowledge graphs in mathematics teaching but also provides valuable insights and methodologies for the development of future AI-driven educational solutions.

Recently, there has been a growing interest in the potential of virtual reality-based teacher training (VRBTT). Despite this surge, the impact of VRBTT on teacher training outcomes remains unclear and cannot be generalized from generic VR-based training approaches. This study aimed to evaluate the overall effects of VRBTT and explore potentially effective instructional design features during the planning, implementation, and evaluation phases of VRBTT, respectively, based on the NLN Jeffries simulation theory. To achieve this, data were collected from Web of Science, Scopus, and Google Scholar. A meta-analysis was constructed based on 58 studies across 29 articles, involving a total of 1421 participants, covering the period from 2013 to 2023. The findings revealed that VRBTT had a low-medium effect on teacher training outcomes. Notably, during the planning phase, three instructional design features of VRBTT—objective, learning design model, and learning loop—significantly moderated the effect sizes. During the implementation phase of VRBTT, three instructional design features—avatar autonomy, static feedback, and participant role—were significant moderators of effect sizes. No significant moderators were found during the evaluation phase of VRBTT. Following the detailed discussion of the findings, this study offered recommendations on instructional design guidelines for VRBTT.