According to the dominant view in cognitive science, language processing requires perceptual simulation of symbols. Various experiments have shown that words that share a perceptual relationship are processed faster. We have proposed an alternative view in which perceptual cues are encoded in language. However, experiments supporting perceptual simulation or language statistics have focused on concept words. It remains therefore unclear whether the evidence found for language statistics might actually just be evidence for perceptual simulations. We presented subjects with lexical items as well as stimuli unlikely to be represented in the perceptual world: grammatical items. Results showed that response times to lexical items could be explained by a statistical linguistic approach and a perceptual simulation approach, supporting both perceptual and symbolic accounts. Results for the responses to grammatical items were explained by statistical linguistic information but not by a perceptual simulation account, raising questions about the principle of parsimony.