Variation and change in a specialized register. A comparison of random and sociolinguistic sampling outcomes in Desert Island Discs.

Corpus-based studies of specialized registers typically sample texts using random methods as far as possible, but they disregard social characteristics of the speakers/writers. In contrast, in corpus-based studies of conversation and quantitative sociolinguistic studies, sampling is more typically designed to optimize social representation. To our knowledge, this study is the first to compare linguistic outcomes from random versus sociolinguistic sampling in a specialized register. Our data comes from the biographical radio chat show, Desert Island Discs (DID), at different points in time. We constructed two versions of a DID corpus: a sociolinguistic judgment sample based on guest demographics, and a random sample. We compare grammatical usage between them using an inductive (‘key POS-tags’) method and close manual analysis, uncovering some evidence of significant grammatical differences between the samples and differing patterns of diachronic change. We discuss the implications of our research for corpus design, representativeness and analysis in specialized registers.