Introduction

Sociolinguists have long grappled with the difficult problem posed by units of analysis, those abstract theoretical compartments into which we fit the ongoing flow of speech and social interaction. At one level, this analytic reification captures those aspects of linguistic interaction that are specifiable in the abstract across contexts, the regularities that can be presupposed before any particular speech act takes place. Yet at another level, an analysis in these terms partially obscures the dynamic relation whereby language and context create and reflect social meaning in spontaneous and unpredictable ways. In addition to the perhaps unavoidable problem entailed by analytic abstraction, an added level of difficulty arises when the analyst's conception of language and its use differs from that of the speakers. This chapter examines the ongoing struggle to capture cross-contextual sociolinguistic regularities, focusing in particular upon the study of code-switching. It draws upon data from a Gaelic—English bilingual community in Canada to demonstrate the tension involved in applying abstract analytic categories to speech, a tension that is heightened when the speakers themselves hold a different, more contextual model of speaking. Although in one sense this tension is inevitable, an adequate linguistic analysis must take account of informants' own metapragmatic categories; to apply the standard sociolinguists' “grid” without generating this account is to risk a distorted view of linguistic processes.