Influence of the tree prior and sampling scale on Bayesian phylogenetic estimates of the origin times of language families

Bayesian phylogenetic methods derived from evolutionary biology can be used to reconstruct the history of human languages using databases of cognate words. These analyses have produced exciting results regarding the origins and dispersal of linguistic and cultural groups through prehistory. Bayesian lexical dating requires the specification of priors on all model parameters. This includes the use of a prior on divergence times, often combined with a prior on tree topology and referred to as a tree prior. Violation of the underlying assumptions of the tree prior can lead to an erroneous estimate of the timescale of language evolution. To investigate these impacts, we tested the sensitivity of Bayesian dating to the tree prior in analyses of four lexical data sets. Our results show that estimates of the origin times of language families are robust to the choice of tree prior for lexical data, though less so than when Bayesian phylogenetic methods are used to analyse genetic data sets. We also used the relative fit of speciation and coalescent tree priors to determine the ability of speciation models to describe language diversification at four different taxonomic levels. We found that speciation priors were preferred over a constant-size coalescent prior regardless of taxonomic scale. However, data sets with narrower taxonomic and geographic sampling exhibited a poorer fit to ideal birth–death model expectations. Our results encourage further investigation into the nature of language diversification at different sampling scales.