Séminaire

Greenberg’s U20:Turning generative approaches into models of language change
 

This talk develops a diachronic perspective on Greenberg’s Universal 20 by turning two influential synchronic analyses – Cinque’s movement-based account and Abels & Neeleman’s mixed base-generation-and-movement account – into explicit models of possible transitions between different nominal word orders. We model grammatical change in terms of minimal featural changes: in the Cinque model, transitions involve the addition or loss of movement-triggering features, while the Abels & Neeleman model also allows changes in head–complement directionality. We then use Bayesian phylogenetic inference on a dataset of 1,732 languages and dialects from 145 language families to estimate the long-term stationary frequencies of the different word-order states while controlling for genealogical non-independence. The resulting model comparison provides strong support for the transition system derived from Cinque’s analysis over the one derived from Abels & Neeleman. At the same time, however, the inferred stationary frequencies do not confirm Cinque’s markedness predictions: less marked orders are not systematically more diachronically stable than more marked ones.