Register as a predictor of linguistic variation

Over the last two decades, corpus analysis has been used as the basis for several important reference grammars and dictionaries of English. While these reference works have made major contributions to our understanding of English lexis and grammar, most of them share a major limitation: the failure to consider register differences. Instead, most reference works describe lexico-grammatical patterns as if they applied generally to English. The main goal of the present paper is to challenge this practice and the underlying assumption that the patterns of lexical-grammatical use in English can be described in general/global terms. Specifically, I argue that descriptions of the average patterns of use in a general corpus do not accurately describe any register. Rather, the patterns of use in speech are dramatically different from the patterns in writing (especially academic writing), and so minimally an adequate description must recognize the two major poles in this continuum (i.e., conversation versus informational written prose). The paper begins by comparing two general corpus approaches to the study of language use: variationist and text-linguistic. Although both approaches can be used to investigate the use of words, grammatical features, and registers, the two approaches differ in their bases: the first gives primacy to each linguistic token, while the second gives primacy to each text. This difference has important consequences for the overall research design, the kinds of variables that can be measured, the statistical techniques that can be applied, and the particular research questions that can be asked. As a result, the importance of register has been more apparent in text-linguistic studies than in studies of linguistic variation. The bulk of the paper, then, argues for the importance of register at all linguistic levels: lexical, grammatical, and lexico-grammatical. Analyses comparing conversation and academic writing are discussed for each level, showing how a general average description includes some characteristics that are not applicable to one or the other register, while also omitting other important patterns of use found in particular registers.