Proceedings of the Annual Meeting of the Cognitive Science Society

What drives word order in transitive sentences? Is the oft-noted universal preference for agents to be in the first free argument position a key factor in shaping word order, or is this agent-first principle a mere epiphenomenon of one or more other preferences that are all to some extent correlated with agentivity, in particular the tendency for human referents to precede nonhuman referents, pronouns to precede full noun phrases, shorter arguments to precede longer arguments and given discourse referents to precede new discourse referents? Corpus evidence from 81 languages confirms the universality of these word order principles across a large and diverse set of languages. Using random forest classification models trained to predict the relative position of an argument, we show that these principles do not equally shape word order preferences and that agentivity indeed outcompetes the other principles, suggesting that it is the primary factor driving word order preferences.