L2 Students' Speaking Skills in Robot‐Assisted Language Learning: A Meta‐Analysis

ABSTRACT Previous studies on the impact of robot‐assisted language learning (RALL) interventions on second language (L2) learners' speaking skills have yielded contradictory results. To clarify the interrelationships among these variables, the present study employed a meta‐analysis to synthesise previous studies. A total of 89 potential studies were screened, with 15 meeting the inclusion criteria based on rigorous methodological standards. The included experimental studies varied considerably in design, sample size, intervention type and measurement tools. To ensure comparability across studies, quantitative outcomes were standardised using Hedges'g, minimising small‐sample bias. Effect size calculations and meta‐regression analyses were conducted using Comprehensive Meta‐Analysis (version 4), with moderators including learners' anxiety, academic level and sample size. RALL has a statistically significant positive effect on L2 speaking skills, with pooled effect sizes demonstrating both consistent and substantial impacts when accounting for heterogeneity across studies. Meta‐regression revealed that higher anxiety levels and larger sample sizes were significant positive predictors of effect sizes, whereas academic level showed no significant influence. These results suggest that while anxiety is often conceptualised as a barrier, it may under certain conditions motivate learners to engage more actively in RALL tasks. The findings highlight both the potential and limitations of RALL, emphasising the need for cautious interpretation given the small number of studies and variability in contexts. Future research should further investigate contextual and demographic moderators, as well as long‐term outcomes, to refine understanding of how RALL can be effectively integrated into L2 speaking instruction.