Little agents of big language change

Abstract There are two general views on the connection between language acquisition and change. Functional usage-based theories usually argue against children as agents of phonological and morphosyntactic changes. By contrast, formal rule-based accounts assume that incrementation of new variants in a linguistic community can be triggered by the acquisition process. In this paper, I argue that the latter point of view is reconcilable with a data-driven approach and is in fact corroborated by probabilistic language modelling. Using extensive corpus data of child speech and a new Bayesian model of collostructional analysis, I investigated how children learn construction-verb relations during the acquisition of English grammatical future. I hypothesize that this process includes the following developmental stages: (i) children learn which words their parents preferably use in certain constructions and reproduce these patterns, (ii) children introduce some innovations into the acquired system, and (iii) once their own preferences are established, children become less sensitive to the input from child-directed speech; as the new system grows more self-contained, further innovations become less likely. The innovation curve constructed by my model resembles the so-called “reverse-U model of incrementation” which suggests that incrementation of new variants in language can have its origin in the acquisition process.