Analyzing linguistic variation using discursive worlds

Researchers in variationist sociolinguistics have long sought to develop social measures that are more sophisticated than demographic categories such as age, gender, and social class, while still being useful for quantitative analysis. This paper presents one such new measure: discursive worlds. For each speaker in a corpus, their discursive world is operationalized through compiling a list of specific referents cited in their interview. These lists are then used to construct similarity spaces locating the speakers along dimensions that are discursively relevant in the corpus. Using common clustering algorithms, the corpus speakers are then partitioned into categories, and this partition can be used in statistical analysis. We show how this method can be used to analyze a series of lexical variables in the Cartographie linguistique des féminismes corpus, a corpus of francophone interviews with feminist and queer activists, for which, we argue, quantitative analysis using classic demographic categories is inappropriate.