In recent years, the integration of technology-enhanced learning in early childhood education has attracted growing attention, being recognised as a promising approach to improving learning outcomes and promoting educational equity. In this context, researchers have increasingly explored how diverse resources such as multimedia, gamified learning, and artificial intelligence (AI) can be effectively integrated into second language acquisition (SLA) processes, particularly among young learners. Within the context of China’s ongoing educational reform, early primary education is experiencing pedagogical innovation, creating heightened demand for AI-powered educational tools. Despite the increasing adoption of AI in educational settings, empirical evidence on its application in early childhood second language (L2) learning remains limited. Prior studies have demonstrated specific advantages of AI tools in providing personalised, interactive learning experiences; however, significant questions exist regarding their effectiveness in enhancing receptive vocabulary development, fostering learner motivation, and supporting enjoyment of learning among young children. Moreover, as central figures in children's early learning environments, parents' and caregivers' perceptions of AI-based applications constitute an essential dimension of educational technology integration that requires further investigation. To address these gaps, the study adopts a mixed-methods approach to explore the effects of a multimodal AI-based language learning application (Zebra AI), which integrates Automatic Speech Recognition (ASR), adaptive feedback, and learning analytics, on English as a second language development among young Chinese learners aged 5 to 7. The research is theoretically grounded in SLA theories, Multimedia Learning Theory (MLT), Self-Determination Theory (SDT), and the Technology Acceptance Model. A quasi-experimental design was employed, supported by post-intervention questionnaires and thematic analysis of qualitative interviews. Participants (N = 85) were children aged 5 to 7 years, recruited from two primary schools in central China and randomly assigned to either an experimental group (using the Zebra AI application) or a control group (using a non-educational entertainment app). Over a 12-week intervention period, children’s receptive vocabulary was assessed using the Peabody Picture Vocabulary Test–Fifth Edition (PPVT-5), while cognitive skills, including working memory (WM) and theory of mind (ToM), were measured through standardised tasks. Data were analysed using mixed-design ANOVA and regression modelling. The quantitative findings revealed significant improvements in receptive vocabulary in the experimental group compared to the control group F (1,83) = 52.42, p <.001, η2p = .387, along with significant gains in WM (p < .001). No statistically significant differences were observed in ToM performance. Regression analysis demonstrated that the model significantly explained 44.8% of the variance in receptive vocabulary growth (R2adj = .448, p < .001). Both the group variable (β = .586, p < .001) and age (β = .240, p = .005) were significant predictors of vocabulary gains. Amongst the core cognitive mechanisms, growth in WM exhibited a marginally significant positive predictive trend (p = .051), whereas the impact of ToM did not reach statistical significance. These findings partially support the hypothesis regarding the relationship between WM and vocabulary acquisition, highlighting the potential contribution of associative learning mechanisms to language development under controlled experimental conditions. Data from questionnaires measuring children’s enjoyment with app-based learning indicated high satisfaction rating for the AI application’s usability, visual appeal, and motivational features. Gamified elements, speech recognition, and interactive animation were identified as key factors in sustaining learner engagement. Thematic analysis of semi-structured interviews with parents and caregivers (N = 8) supported these results, revealing observed improvements in children’s vocabulary, motivation, and self-directed learning. Parents and caregivers also reported advantages such as flexibility, reduced homework supervision stress, and cost-effectiveness, while expressing concerns about screen time, device compatibility issues, inappropriate content, and low digital literacy. They suggested future design should support offline access, dialects, peer-support and parent-caregiver training to enhance usability in diverse early childhood learning contexts. This study contributes to the expanding literature in Intelligent Computer-Assisted Language Learning (ICALL), demonstrating empirical evidence for the effectiveness of AI-enhanced tools in early SLA education through the specific application of Zebra AI. However, it is important to note that these findings are specific to the algorithmic and pedagogical design of the tested application and may not be directly applied to other AI tools with different functions. It confirms this technology’s capacity to enhance both learning outcomes and engagement, offering insights for educators, designers, and policymakers within the context of specific design features, to promote effective, equitable early language learning in digital learning environments.
Jiachen Wang (Thu,) studied this question.