Human language contains regular syntactic structures and grammatical patterns that should be detectable in their co-occurence networks. However, most standard complex network measures can hardly differentiate between co-occurence networks built from an empirical corpus and a body of scrambled text. In this work, we employ a motif extraction procedure to show that empirical networks have much greater motif densities. We demonstrate that motifs function as efficient and effective shortcuts in language networks, potentially explaining why we are able to generate and decipher language expressions so rapidly. Finally we suggest a link between motifs and constructions in Construction Grammar as well as speculate on the mechanisms behind the emergence of constructions in the early stages of language acquisition.