Generative AI has repositioned English as a foreign language (EFL) teachers from technology consumers to architects of AI-driven teaching innovation (AITI). Yet, little is known about the psychological mechanisms that determine whether this transition succeeds. Grounded in Self-Determination Theory (SDT) and the Ecological Model of Teacher Agency, this explanatory sequential mixed methods study tested whether AI anxiety and teacher agency sequentially mediate the relationship between AI literacy and AITI among EFL teachers. In the quantitative phase, 361 EFL teachers completed validated self-report measures, analyzed through structural equation modeling (SEM). The findings indicated that AI literacy positively predicted AITI directly and indirectly, with AI anxiety and teacher agency emerging as significant sequential mediators. In the subsequent qualitative phase, semi-structured interviews with 21 teachers, analyzed through reflexive thematic analysis, both substantiated and extended these pathways. Findings revealed that anxiety was relationally produced through professional comparison, that agency functioned as deliberate boundary-setting around AI’s instructional jurisdiction, and, most critically, that this psychological pathway unfolded largely in the absence of institutional policy or support, leaving teachers to innovate alone. The integrated findings suggest that cultivating AI literacy is necessary but insufficient; sustained innovation requires institutional scaffolding alongside individual psychological readiness.
This article investigates how integrated configurations of critical thinking dispositions, combining self efficacy, epistemic beliefs, and reasoning style, influence educational video evaluation and school related usage decisions among 1,079 secondary school students. Fuzzy C-Means clustering identified three distinct dispositional profiles at each educational level. In high school, two high disposition subtypes emerged: an intuitive profile (high self efficacy, low analytical reasoning) and an analytical profile (moderate self efficacy, high analytical reasoning). Only the analytical profile linked to superior critical performance and confirmation bias resistance. The intuitive profile, despite high confidence, matched low disposition students. In middle school, performance scaled with overall disposition intensity, without reasoning style moderation. Findings indicate that effective high school critical thinking requires motivation analytical reasoning coupling, suggesting a late adolescent threshold. The study informs differentiated pedagogy and computer-based decision support systems (e.g., learning analytics platforms) in digital education.
The rapid development of educational robots (ERs) and their introduction in various educational fields have made the topic of ERs increasingly important, especially in K-12 education. This study aims to provide educational technology researchers and teachers with a more comprehensive understanding of the research focus and the application of ERs in K-12 technology-assisted learning environments. Adopting a technology-based learning model and using the Web of Science (WoS) database as the primary data source, this study systematically reviewed the application issues, research issues, and interaction issues of ERs in K-12 (ER-K12) publications in education science and technology SSCI journals from 2005 to 2024. We further analyzed the ER roles, distribution of types, and learning strategies used in different application domains. The review results showed that ER-K12 research is growing rapidly, with science and programming as the main subjects, but research in other subjects, such as language and arts, is also increasing. After cross-comparison analysis, this study found that there are clear differences in the types of robot applications, roles of robots, and learning strategies adopted in different subjects using ERs. Science and programming prefer technology-oriented applications, where ERs are often used as tools with structured and inquiry-oriented strategies. Language and arts, on the other hand, emphasize interaction and creative expression, and ERs often take on the roles of instructors and peers, with contextual learning and digital storytelling. Based on the above findings, this study also suggests future research directions for researchers and teachers.
This study investigates the use of local Large Language Models (LLMs): Gemma, LLama, Mistral Small, and Phi, in assessing Spanish second language (L2) writing. While commercial models like ChatGPT have already been used for this task, they present privacy risks and reproducibility challenges. To evaluate local alternatives to ChatGPT, human raters were used to assess a third-semester assignment, evaluating 100 paragraphs to provide a baseline of comparison for the local LLMs’ assessment. Results showed varying performance: while some models approached the human baseline; others were inaccurate and inconsistent. On average, they performed better using a holistic rubric than an analytic one. A third experiment, averaging the top 3 models: Llama3.3:70b, Gemma2:9b, Mistral-Small:22b, achieved the highest agreement (r = 0.88), surpassing inter-human rating. A separate experiment testing RAG and fine-tuning techniques on the smallest model: Llama3.1:8b, showed no significant improvements with RAG but substantial gains in accuracy and consistency with fine tuning. Findings suggest that local LLMs show promise for Spanish L2 writing assessment. Local LLMs are preferable to proprietary ones because they protect privacy and reduce risks of data exposure, they may better support replicable workflows and may offer practical advantages for long term adoption.
With the rapid development of generative artificial intelligence (GenAI), leveraging its capabilities to optimize video learning has emerged as a critical research topic. However, empirical evidence about its effectiveness and underlying cognitive mechanisms remains scarce. This study recruited 120 undergraduate and postgraduate students and employed multimodal analysis to investigate the impact of a GenAI-supported questioning mechanism (AI-VL mode) on learners’ intrinsic motivation, cognitive load, attention allocation, and learning performance. Results showed that the AI-VL mode significantly outperformed traditional video learning (TVL) across intrinsic motivation, cognitive load, and learning performance, and further outperformed embedded question-and-answering video learning (QA-VL) specifically in learning performance. Regarding cognitive processing, eye-tracking metrics confirmed that the AI-VL mode redistributed cognitive resources through real-time feedback, reducing intrinsic and extraneous cognitive load. Furthermore, Lag Sequential Analysis of interactive behaviors revealed a cross-modal attentional cycle, specifically the “feedback-viewing” path, which enabled the reallocation of liberated cognitive resources into germane cognitive load. However, cluster analysis revealed that most learners remained at a surface-level interaction stage. These findings provide empirical evidence and practical guidance for designing adaptive GenAI-driven educational scaffolds for future online video learning platforms.
