Although motor activation during action-word-processing has received considerable attention, the effects of the linguistic context have largely been overlooked. To assess how flexible and context dependent motor activation during language processing may be, we experimentally tested the impact of volition modality (want to do) on this activation: volition modality sets an action in a irrealis perspective, since to want to do X presupposes that X is not currently being done. Our experimental design relied on a novel experimental technique developed in our lab (Frak et al., 2010; Aravena et al., submitted), using a grip-force sensor (ATI mini-40) to measure online the effects of language processing on motor behaviour. Participants held the grip-force sensor with closed eyes throughout the experiment while listening to orally presented French action and non-action words in affirmative vs. volitive sentences. Relative to non-action words a significant enhancement of grip force was observed for action words in declarative sentences starting around 250ms after target word onset. In volitional contexts, however, the same action-words elicited a significantly reduced grip-force amplitude. Our results clearly demonstrate that motor brain structures are not activated mandatorily during the processing of action words; motor activation is modulated by the linguistic context and more specifically, the manner in which the corresponding action concept is recruited. Consequences for models of embodied semantics are discussed. Our findings further confirm that our simple experimental paradigm can advantageously be used for illuminating online studies of the crosstalk between language and the motor systems that are also ecological and economical.