Research and thinking into the cognitive aspects of language evolution has usually attempted to account for how the capacity for learning even one modern human language developed. Bilingualism has perhaps been thought of as something to think about only once the ‘real’ puzzle of monolingualism is solved, but this would assume in turn (and without evidence) that bilingualism evolved after monolingualism. All typically-developing children (and adults) are capable of learning multiple languages, and the majority of modern humans are at least bilingual. In this paper I ask whether by skipping bilingualism out of language evolution we have missed a trick. I propose that exposure to synonymous signs, such as food and alarm calls, are a necessary precondition for the abstracting away of sound from referent. In support of this possibility is evidence that modern day bilingual children are better at breaking this ‘word magic’ spell. More generally, language evolution should be viewed through the lens of bilingualism, as this is the end state we are attempting to explain.