Language is a uniquely human trait but has to interface with a host of pre-existing abilities, a requirement that might have shaped certain properties of language.  I will present two case studies of such pre-existing mechanisms: a sensitivity to positional memory for sequences and a sensitivity to identity relations. I show that these simple, evolutionarily ancient mechanisms shape certain universal linguistic structures, allow nonhuman animals to learn similar structures, and predict which cues humans can and cannot use to learn words from fluent speech. Further, I show that, by focusing on the limitations of these mechanisms, we can refine our hypotheses about the properties of specifically linguistic computations. The language faculty might thus have recycled certain pre-existing mechanisms for its own, linguistic, purposes; while the properties of these mechanisms seem to constrain structural properties of language, their limitations give some glimpses of what might be more specific to linguistic computation.