Tracking the change that leads to typological variation

Word order universals and the phylogenetic comparative method

Abstract

Subject-object-verb (SOV) and subject-verb-object (SVO) word order are the most common word orders across the languages of the world, jointly covering more than 80% of the documented languages. To what degree this biased distribution reflects genuinely universal tendencies of language change, as opposed to historical contingencies, is an open question. The current distribution is shaped by at least three confounded factors: the inherent directionality of language change, differential diversification rates of language families with different typological profiles, and the properties of ancestral languages inherited by their extant descendants. Plain counting of the current distribution cannot disentangle these factors. I deploy the phylogenetic comparative method from computational biology and model word-order change as a continuous-time Markov chain (CTMC) running on language phylogenies inferred via Bayesian phylogenetic inference. The language sample comprises 1,567 languages: 1,443 from 107 multi-member families and 124 singleton lineages; family phylogenies were inferred from lexical data using topological constraints from Glottolog and time calibrations from a Bayesian global tree. The fitted model assigns an equilibrium probability of 65% to SOV and 28% to SVO, clearly diverging from the apparent symmetry between the two types. The dominant SOV/SVO exchange is asymmetric, with a net flux from SVO toward SOV.