According to embodied theories, motor and language processing bidirectionally interact: Motor activation modulates behavior in lexico-semantic tasks (semantic resonance), and understanding motor-related words entails activation of the corresponding motor brain areas (motor resonance). Whereas many studies investigated such interaction in the first language (L1), only few did so in a second language (L2), focusing on motor resonance. Here, we directly compared L1 and a late L2, for the first time both in terms of semantic and motor resonance and both in terms of magnitude and timing, by taking advantage of single-pulse TMS. Twenty-five bilinguals judged, in each language, whether hand motor-related (“grasp”) and non-motor-related verbs (“believe”), were physical or mental. Meanwhile, we applied TMS on the hand motor cortex at 125, 275, 350, and 500 msec post verb onset, and recorded behavioral responses and TMS-induced motor evoked potentials. TMS induced faster responses for L1 versus L2 motor and nonmotor verbs at 125 msec (three-way interaction β = −0.0442, 95% CI [0.0814, −0.0070]), showing a semantic resonance effect at an early stage of word processing in L1 but not in L2. Concerning motor resonance, TMS-induced motor evoked potentials at 275 msec revealed higher motor cortex excitability for L2 versus L1 processing (two-way interaction β = 0.095, 95% CI [0.017, 0.173]). These findings confirm action–language interaction at early stages of word recognition, provide further evidence that L1 and L2 are differently embodied, and call for an update of existing models of bilingualism and embodiment, concerning both language representations and processing.