The Impact of Pedagogical Agents on Second Language Acquisition Outcomes: A Meta-Analysis Guided by Social Agency Theory

Digital Pedagogical Agents (DPAs) are virtual teaching entities created in the digital learning environment that have characteristics such as emotion and socialization. According to the basis of Social Agency Theory (SAT), a three-level random-effects meta-analytic model is employed in this paper to study the general effect and boundary conditions of DPAs in Second Language Acquisition (SLA) environments systematically. Through searches across seven Chinese and English databases, 31 independent empirical studies were identified, yielding 112 effect sizes (2002–2026). Based on the three-level random-effects model, the effect of DPA on SLA outcomes is about 0.607 (Hedges’ g = 0.607; 95% CI [0.45, 0.77]; p < 0.001), and there is a relatively large between-study heterogeneity. According to the moderation analysis, the magnitude of the effect varies greatly with different kinds of study designs, cultural contexts and learner characteristics, and although there is a moderating effect by publication type, it is not statistically significant. Funnel plot inspection, Egger’s test, and the trim-and-fill method all showed the presence of publication bias, yet the adjusted effect size remained robust (Hedges’ g = 0.521). This study provides quantitative evidence for understanding the mechanisms of DPAs in SLA and offers practical implications for instructional design and technological development in intelligent language learning environments.