Language vs individuals in cross-linguistic corpus typology

Abstract

There is a long tradition in linguistics of seeing each language as a powerful factor setting out predetermining grooves in how people express themselves. But how strong is this effect? We know that despite the forces of linguistic habit people nonetheless enjoy some freedom in formulating their thoughts. Can we measure the relative contributions of language structures and individual variation to how people formulate statements about the world? Do accounts of typological differences need to take individual variation into account, and is such variation more prevalent in some kinds of linguistic domains than others? In this paper, we deploy a parallax corpus across thirteen languages from around the world and explore four case studies of linguistic choice, two grammatical and two semantic. We assess whether differences are accounted adequately just by individual participant variation, just by language information, or whether taking into account both helps account for the patterns we see. We do this through comparisons of statistical models. Our results make it clear that participants using the same language do not always behave similarly and this is especially true of our semantic variables. We take this to be a strong caution that the behaviour of individual participants should be considered when making typological generalisations, but also as an exciting outcome that corpus typology as a field can help us account for intra- and inter-language variation.