Advances in generative artificial intelligence (GenAI) have the potential to transform learning. As research increasingly calls for transforming the human-GenAI relationship from mere tool use to a collaborative partnership, there remains a notable gap in exploring an important factor influencing this relationship: trust in GenAI, being crucial amidst growing concerns about disuse, misuse, and abuse of GenAI in learner-GenAI interaction primarily via language. This study seeks to address the gap by investigating university students’ trust in GenAI, its related factors of reliance and resistance, and boundary conditions in using GenAI for language learning. The research utilized an explanatory sequential mixed-methods design. In Study 1, a proposed conceptual model within the stimulus-organism-response framework was quantitatively tested through structural equation modeling using a survey. Data were collected from 682 university students who used GenAI in language learning. Study 2 further confirmed the results through 40 qualitative interviews and refined the conceptual model by offering a deeper understanding of the relationships. The findings indicate that trust in GenAI is a significant factor in university students’ use of GenAI for language learning, with its influence on behavioral intention to use mediated through reliance and resistance. Moreover, forms of the five factors are identified: trust in GenAI, reliance, resistance, perceived risk, and actual use. Additionally, two boundary conditions are identified: perceived risk between trust in GenAI and reliance/resistance, and self-directed learning between behavioral intention to use and actual use. These findings provide theoretical, research, and practical implications regarding the appropriate use of GenAI in education.

As learners are able to perceive interactivity when interacting with instructors or peer learners in traditional learning environments, learners are similarly able to perceive interactivity when interacting with artificial intelligence (AI) in AI-supported learning environments. Advancements in AI, such as generative AI including ChatGPT and Midjourney, enhance learners’ perceived interactivity, thereby facilitating learning through AI-enabled interaction. However, there is no scale in education for measuring perceived interactivity of learner-AI interaction. This study develops a 17-item scale to assess the extent to which a learner perceives interactivity with AI from four dimensions: responsiveness, personalization, learner control, and learning engagement. The sample group included 422 Chinese university students for the first application and 306 university students for the second application. Both the exploratory factor analysis and the confirmatory factor analysis verified the factor structure of the scale. The Cronbach’s alpha value for the whole scale was 0.948, whereas the Cronbach’s alpha values for the four dimensions ranged between 0.820 and 0.915. Results suggested that this scale was a reliable and valid instrument. This study also found that perceived interactivity of learner-AI interaction was significantly associated with AI tools, learners’ behavioral intentions to use AI in learning, months of using AI in learning, and average duration of using AI in learning each time, and not associated with ages, genders, education levels, and fields of education. Finally, theoretical and practical implications are discussed.

This quantitative synthesis included 48 qualified studies with a total sample of 116,051 K-12 students. Aligned with previous meta-analyses, the findings suggested that formative assessment generally had a positive though modest effect (ES = + 0.19) on students’ reading achievement. Meta-regression results revealed that: (a) studies with 250 or less students yielded significantly larger effect size than large sample studies, (b) the effects of formative assessment embedded with differentiated instruction equated to an increase of 0.13 SD in the reading achievement score, (c) integration of teacher and student directed assessment was more effective than assessments initiated by teachers. Our subgroup analysis data indicated that the effect sizes of formative assessment intervention on reading were significantly different between Confucian-heritage culture and Anglophone culture and had divergent effective features. The result cautions against the generalization of formative assessment across different cultures without adaptation. We suggest that effect sizes could be calculated and intervention features be investigated in various cultural settings for practitioners and policymakers to implement tailored formative assessment.