A series of results from event-related brain potential recordings and fMRI research will be presented, suggesting that language processing does not obey strict compositionality, and, moreover immediately recruits extralinguistic information. It will also be shown that pragmatic inferences require contributions from TOM networks. This implies that an embodied account of semantics fails (under the somewhat strange assumption that the brain is not part of the body). I will put forward an embrained perspective on language processing.