08.12.2026 Tim Vaughan

While the application of computational methods to the problem of linguistic phylogeny reconstruction is not new, recent years have seen a substantial increase in the use of model-based Bayesian approaches, in many cases originally developed with biological or epidemiological applications in mind.  These approaches excel at jointly inferring phylogenies and parameters of interest, such as speciation and extinction rates, while properly accounting for statistical uncertainty.

Although these model-based approaches are useful, they can be complex to apply in practice. In this talk I will describe three ways in which we have been tackling some of these practical challenges.

I will begin by considering the problem of accounting for non-treelike diversification patterns arising from the borrowing of words between languages.  Such patterns break the fundamental assumption of phylogenetic inference and can lead to severe biases if ignored.  I will present two new approaches aimed at accommodating non-vertical inheritance in phylolinguistic analyses.

Secondly I will discuss the application of so-called "calibration priors" in phylogenetic inference to calibrate the time scale of language phylogenies. While this common practice is widely understood to be theoretically problematic, I will demonstrate several ways in which it can be done correctly.

Finally, I will briefly present PhyloSpec: a new language for formulating and communicating phylogenetic and phylodynamic models in an platform-agnostic manner.  Integration of this modeling language into BEAST2 and other phylogenetic inference systems promises to improve transparency and replicability of Bayesian phylodynamic analyses.