Balanced Samples and Imagined Populations

This chapter critically examines representativeness in corpus linguistics, challenging assumptions about balanced corpora and their ability to represent language varieties. The author distinguishes between three types of ‘representativeness’: random sampling of language as a whole, representation of specific text types, and representation of speaker populations. Using modal verb variation across genres in the Diachronic Corpus of Present-day Spoken English (DCPSE), the chapter demonstrates that linguistic phenomena are often more sensitive to genre than previously recognised. The chapter argues against assumptions that balanced corpora with consistent sampling frames can represent ‘English in general’. Instead, the author proposes ‘imagined populations’ where the realistic generalisation scope is limited to infinite collections of comparable corpora collected under similar conditions and proportions. Analysis reveals that pooling data across genres can obscure important variation, as averages may conceal significant differences between text types, evidenced by contrasting patterns in parliamentary language versus legal cross-examination. Methodologically, the chapter advocates explicit reporting of sampling processes and careful consideration of genre effects. For single-genre samples, claims should be restricted to that specific text type. For multi-genre samples, results should be understood as generalisable only to imagined populations of similar corpora, not language varieties generally. The chapter recommends collecting many short texts rather than few long ones to maximise participant diversity, and analysing sub-genres separately when sufficient material is available.