Abstract
Embodied and grounded cognition theories have assumed that the sensorimotor system is causally involved in processing motor-related language content. Although a causal proof on a single-cell basis is ethically not possible today, the present fMRI study provides confirmation of this longstanding speculation, as far as it is possible with recent methods, employing a new computational approach. More specifically, we were looking for common activation of nouns and objects, and actions and verbs, representing the canonical and mirror neuron system, respectively. Using multivariate pattern analysis, a resulting linear classifier indeed successfully generalized from distinguishing actions from objects in pictures to distinguishing the respective verbs from nouns in written words. Further, these action-related pattern responses were detailed by recently introduced predictive pattern decomposition into the constituent activity atoms and their relative contributions. The findings support the concept of canonical neurons and mirror neurons implementing embodied processes with separate roles in distinguishing objects from actions, and nouns from verbs, respectively. This example of neuronal recycling processing algorithms is consistent with a multimodal brain signature of human action and object concepts. Embodied language theory is thus merged with actual neurobiological implementation.