Recursion has become a lamp for the linguistic moths – it has become an obsession far from the centre of what linguistics should be focused on. It plays a limited role in the structure of many languages, indefinite recursion is of course never actually displayed, and what is exemplified could therefore always be modeled in practice by finite state devices. There are many more central puzzles to focus on, like the diverse specific structures mapped on strings, rather than the mechanisms that generate unstructured string-sets. Embedded clauses have been the main focus of interest, but it is noteworthy that (a) many languages offer very limited embedding possibilities; (b) some which do have embedding effectively cap embedding at one deep; (c) almost any such embeddings can be paraphrased by parataxis (strings of adjoined clauses as in veni, vidi, vici). Parataxis is why many languages can lack embedded clauses of different kinds without any loss of expressive power: the expressive power is always present in the pragmatics whether or not it is there in the syntax. To make the point that expressive power lies in the pragmatics, I’ll examine centre-embedding in interactive discourse. Centre-embedding has the virtue that it is easily distinguished from parataxis – which is not the case for edge-recursion in many languages. Centre-embedding in clauses is effectively capped at two deep in all spoken languages (very occasionally three deep in written), apparently by memory and parsing limitations. But centre-embedding in interactive discourse can break this barrier, and does so routinely. The explanation for this is actually unclear, but the phenomenon would seem to show the advantages of distributed cognition. Rather than thinking of recursion as the performance-limited “externalization” of an individual competence, the discourse phenomena suggest that interactive language usage, where centre-embedding is hyper-trophied, is the natural home base and the ultimate source of complex recursion in the grammatical system.