AI-assisted speaking practice for elementary EFL learners: an action research study in a rural context
TL;DR
The quantitative results showed a statistically significant increase in students’ speaking scores after the instructional period and the qualitative findings suggested that students perceived the activities positively and reported greater willingness to practice, more independent learning behaviors, and a perceived reduction in speaking-related nervousness during classroom activities.
Abstract
Artificial intelligence (AI)-assisted technologies are increasingly used in language education to provide individualized practice and feedback. However, limited research has examined how AI-assisted tools support young English as a Foreign Language (EFL) learners’ speaking practice and self-learning capacity, particularly in rural elementary contexts. This classroom action research study examined the implementation of Microsoft Reading Coach and NaturalReader in a 12-week AI-supported speaking program at a rural elementary school in Taiwan. Thirteen students from Grades 4 to 6 participated in weekly activities focusing on pronunciation, fluency, and autonomous practice. Data were collected through Cambridge A1 Movers speaking pre- and post-tests, weekly Microsoft Reading Coach records, classroom observations, and semi-structured interviews. The quantitative results showed a statistically significant increase in students’ speaking scores after the instructional period. Older students appeared to engage more consistently with the AI-supported routines, while younger students required more scaffolding, modeling, and teacher guidance. The qualitative findings suggested that students perceived the activities positively and reported greater willingness to practice, more independent learning behaviors, and a perceived reduction in speaking-related nervousness during classroom activities. However, because the study used a single-group action research design, the findings should be interpreted as context-specific classroom evidence rather than causal evidence of AI tool effectiveness.