Modelling in the midst of a replication crisis for embodied action semantics: a commentary on Carriere et al. (2026)

A substantial body of research has demonstrated that the motor system is involved in language processing, prompting questions about its role. Carriere et al. (2026) revisit the claim that the motor cortex is engaged in action word and action-sentence processing in a somatotopic fashion, then offer a neural network model simulation of how semantic understanding emerges during word learning based upon this framework. However, the utility of any “brain constrained” model is dependent on the neurobiological plausibility of the effects it is designed to simulate. In this commentary, I argue that multiple sources of evidence, including failures to replicate key findings, demonstrate that semantic processing of action words and action-sentences is unlikely to rely on sensorimotor simulation. Embodied language researchers need to acknowledge the lack of reliable evidence for the semantic somatotopy account and develop alternative models of motor system involvement in language processing that are grounded in reliable effects.