The paper reports an electrophysiological (EEG) study investigating how language is involved in perception-action relations in musically trained and untrained participants. Using an original backward priming paradigm, participants were exposed to muted point-light videos of violinists performing piano or forte nuances followed by a congruent vs. incongruent word. After the video presentation, participants were asked to decide whether the musician was playing a piano or forte musical nuance. EEG results showed a greater P200 event-related potential for trained participants at the occipital site, and a greater N400 effect for untrained participants at the central site. Musically untrained participants were more accurate when the word was semantically congruent with the gesture than when it was incongruent. Overall, language seems to influence the performance of untrained participants, for which perception-action couplings are less automatized.