‘Everyone knows that language is variable’, said Edward Sapir in 1921. However, throughout the history of linguistics, linguists have tended to act as if language were not variable. Most linguistic theories have started from the assumption that variability in language is unmanageable, or uninteresting, or both. Consequently, there has been a tendency to abstract away from the variable data that linguists inevitably encounter in order to begin the analysis at some more homogeneous ‘level’. The analysis of linguistic variability is much more recent, and more and more linguists are coming to see that variability is not only interesting but also that it can be made manageable and integrated into linguistic theory. The main impetus has come from urban dialectologists, and the movement has gradually been joined by mathematical linguists who see linguistic variability as a testing ground for probability theory, by sociologists of language who meet complex variability situations in language planning and multilingual literacy programmes, by linguistic philosophers who are seeking to model variability with many-valued logics and ‘fuzzy’ categories, and, perhaps belatedly considering their long confrontation with variability, by dialect geographers. This chapter outlines some of the main thrusts in the attempt to integrate the analysis of variability into linguistic theory.

The variable as a structural unit

A fundamental paradox of linguistic theory is summed up in the question posed by Uriel Weinreich in the title of a well-known article: ‘Is a structural dialectology possible?’