Over the last ten years, historical linguists have paid increasing attention to multilingualism and multilingual practices, defined as the alternating use of resources from two or more languages by a writer within a single text. Most studies in this emerging field have been carried out using relatively small datasets, such as individual genres or texts written by a single person, which allow only limited opportunities for the discovery of broader tendencies. Furthermore, previous studies have typically examined multilingual practices in light of individual factors without attempting to identify significant predictors in a multifactorial setting. With the 34-million-word multigenre Corpus of Late Modern English Texts 3.0 as our primary data we explore the sociolinguistic and textual factors that predicted the use of foreign languages in English texts during the Late Modern period. To identify the most significant variables, we use recursive partitioning, a statistical method in which significance-based splitting is carried out in a stepwise fashion on predictor variables. The decision trees produced by recursive partitioning are easily interpreted and they allow rule-based predictions about the likely frequency of the linguistic feature being investigated.