Usage-based evolutionary models reveal context-specific word order change in Indo-European

This paper introduces a new corpus-based approach for studying word order change and reconstruction with Bayesian computational phylogenetic methods. We investigate the rates of change in object-verb order in 46 sentences from 36 Indo-European languages extracted from a parallel corpus. A Gaussian mixture model reveals that the rates can be grouped into three components representing syntactic constructions with distinct diachronic dynamics. Contexts with nominal objects are relatively stable, whereas object-verb order in contexts with pronominal objects evolves fast. Complement clauses have a strong diachronic bias towards VO . Stochastic character mapping suggests a VO order of nominal objects and verb in Proto-Indo-European, while the fast rates in contexts with pronominal objects do not allow a reliable reconstruction of ancestral states.