AI-driven mobile practice and English-speaking performance: a quasi-experimental study of ELSA speak among Malaysian university learners

Abstract

This study examined whether supplementary use of the AI-enabled ELSA Speak application was associated with improved English-speaking performance among Malaysian university learners. A mixed-methods, quasi-experimental pre-test–post-test non-equivalent control-group design involved 60 undergraduates from two intact sections of a compulsory English course. Both sections received identical core instruction; the experimental section also completed approximately 36 hours of monitored ELSA Speak practice over four months, scheduled three times weekly for at least 30 minutes. Across approximately 48 scheduled sessions, cumulative records implied about 45 minutes per session rather than a recorded session-level mean. An IELTS-informed analytic rubric assessed pronunciation, fluency, vocabulary, grammatical accuracy, and communicative content. Both sections improved, but the experimental section showed markedly higher post-test performance after accounting for baseline differences. Exploratory interviews and classroom observations suggested learners valued practice and immediate feedback, but did not provide validated psychological or technology-related measures. The advantage cannot be attributed solely to software because additional practice, automated feedback, monitoring, technical guidance, and instructional support were also provided. Findings are context-specific and hypothesis-generating, warranting stronger allocation procedures, a time-matched active comparison, multiple blinded raters, larger samples, and delayed follow-up.