Methods and insights from statistical physics are finding an increasing variety of applications where one seeks to understand the emergent properties of a complex interacting system. One such area concerns the dynamics of language at a variety of levels of description, from the behaviour of individual agents learning simple artificial languages from each other, up to changes in the structure of languages shared by large groups of speakers over historical timescales. In this Colloquium, we survey a hierarchy of scales at which language and linguistic behaviour can be described, along with the main progress in understanding that has been made at each of them---much of which has come from the statistical physics community. We argue that future developments may arise by linking the different levels of the hierarchy together in a more coherent fashion, in particular where this allows more effective use of rich empirical data sets.