Introduction

Like most human activity, language does not fit neatly into the analytic boxes that observers often use to segment, categorise, and theorise about the subject. Whether those boxes are called features, phonemes, or syntactic structures, or rules, constraints, or principles, the facts of language always slop over the edges or ooze from one into another. The customary approach in linguistics is to treat this mismatch between categories and facts as ‘linguistic variation’ – but we should be clear that doing so effectively privileges the analytical categories over the empirical substance. Variation, as traditionally understood, involves single categories being mapped onto variable realisations, as if the categories were primary and given – platonic ideals existing on a higher, purer, plane, that are only imperfectly reflected in the muddy reality of speech. An alternative view, in which natural language in all its richly variegated glory is primary, and the analytical categories are as yet imperfect theoretical constructs that provide only a crude model of reality, is rarely considered. As a healthy terminological corrective, perhaps linguists should consider thinking about variation as highlighting the problem of ‘theoretical inadequacy’.

Nowhere is this lousy fit between theoretical models and variable facts more evident than in the treatment of language change. Since Saussure, linguistic theory has for the most part assumed the irrelevance of diachrony in the construction of formal theory, producing as a consequence static models that not only fail to accommodate change, but actually appear to exclude it as a logical possibility.