Konstantin Kougioumtzis
Embodied cognition has emerged as a prominent theoretical orientation in STEM education research. Although STEM disciplines differ in their epistemic demands, they share a common reliance on abstract, relational, and spatially structured reasoning; features that make them particularly relevant to embodied approaches. The field, however, remains conceptually fragmented: different theoretical traditions, learning mechanisms, and implementation formats are frequently conflated, limiting both theoretical integration and instructional translation. This integrative review synthesizes published review articles, meta-analyses, and theory-oriented syntheses to address three interrelated questions: (a) how is embodied cognition conceptualized in the review-level literature, (b) what learning mechanisms are attributed to embodiment, and (c) under what conditions do these mechanisms operate? The review identifies four partially overlapping theoretical traditions (grounded/simulation-based, enactive/ecological, extended/embedded, and pluralist) and four mechanism clusters gesture-based representation, perceptual–spatial structuring, cognitive offloading, and socially mediated interaction. Drawing on recent meta-analyses, the review indicates that embodied approaches yield moderate positive effects on learning performance but with substantial heterogeneity. Key moderators include learner age, subject domain, embodiment type (active vs. passive), degree of bodily engagement, andthe functional alignment between bodily activity and learning goals. Embodied approaches appear to support learning principally when they supplant rather than compound cognitive demands; effects tend to diminish or reverse under functional misalignment or excessive task complexity. The review concludes that further progress depends less on empirical accumulation than on conceptual clarification: specifying which mechanisms a given embodied interaction activates, for which learners, and under what conditions.