Most studies of genitive variation in English have considered only the choice of two variants ('s versus of), based on analysis of only tokens that are judged to be interchangeable. We argue in the present article that research on genitive variation can be usefully extended in both respects: including premodifying nouns as a third variant; and attempting to account for all tokens of the genitive. In addition, we extend the scope of analysis to explore the possibility of contextual constraints having different importance in different registers.

First, we carry out a text-linguistic analysis comparing the rates of genitive variants in texts from three registers (conversation, newspaper reports, academic articles), showing that genitives overall are much more frequent in written registers, with the premodifying noun variant being especially frequent. Then, a variationist analysis is undertaken to account for the choice of genitive variant in particular contexts and registers. A total of 3,425 genitive tokens were coded for ten contextual characteristics (e.g. length of the Modifying NP, semantic category of the Modifying noun and the Head noun, final sibilancy of the Modifying noun). Statistical analyses with random forests and conditional inference trees are triangulated, showing how contextual factors interact in predicting the use of each genitive variant – and how patterns of variation differ across registers.