According to embodied cognition theory, speech is largely based on the body motor and sensory experience. The question that is crucial for our understanding of the origin of language is how our brain transforms sensory-motor experience into word meaning. We have developed an auditory-motor experimental procedure that allowed investigating neural underpinning of word meaning acquisition by way of associative "trial-and-error" learning that mimics important aspects of natural word learning. Participants were presented with eight pseudowords; four of them were assigned to specific body part movements during the course of learning – through commencing actions by one of a participant’s left or right extremities and receiving a feedback. The other pseudowords did not require actions and thus were used as controls. Magnetoencephalogram was recorded during passive listening to the pseudowords before and after the learning. The cortical sources of the magnetic evoked responses were reconstructed using distributed source modeling. The learning of novel word meaning through word-action associations selectively increased neural specificity for these words in the auditory parabelt areas responsible for spectrotemporal analysis, as well as in articulatory areas, both located in the left hemisphere. The extent of neural changes was linked to the degree of language learning, specifically implicating the physiological contribution of the left perisylvian cortex in the speech learning success.