Recent neurophysiological studies on canonical neurons demonstrate that perceiving manipulable objects automatically activates the same neural circuits that fire during the actual manipulation or grasping of objects: these findings allow us to conceive affordances as «possibilities for action» brought about by objects' physical properties. The paper describes an attempt to apply this notion of affordances to the linguistic domain. Affordances are annotated in a "corpus" focused on Italian grasp verbs, together with semantic features. Classification experiments are conducted in order to distinguish between different senses of verbs (concrete vs. metaphorical). Results show that pragmatic information (affordances) is more relevant than semantic information in assigning the correct interpretation to sentences.