Bayesian phylogenetic dating of language families

node

The application of the computational machinery built for evolutionary biology to cognate data, and since 2003 the most productive and most contested source of absolute dates in historical linguistics. The input is a matrix of cognate judgements over basic vocabulary, produced by ordinary comparative work; the model treats cognate gain and loss as a stochastic process along the branches of an unknown tree, allows rates to vary between branches, and uses historically attested languages and dated extinctions as calibration points. The output is not a single tree but a posterior distribution over trees and dates, which is a genuine methodological advance: uncertainty is reported rather than hidden. The objections are equally genuine. Results are sensitive to the cognate coding, to which languages are sampled, to the calibration priors, and to whether borrowing is modelled at all — and comparative linguists have repeatedly shown that changing defensible inputs moves the headline date by millennia. The technique is best read as a way of testing whether a proposed history is compatible with the lexical data, rather than as an instrument that measures the age of a family.

A node is not a place. Drawing it on a map would assert something about the world that no stored fact supports.