Philipp Rönchen: Evaluating models of vocabulary evolution for linguistic dating and subgrouping

Abstract

The majority of the world’s languages cluster into families such as Indo-European and Austronesian, but the deep history and structure of language families are often poorly understood. Evolutionary models of vocabulary change have the potential to shed light on these questions, yet the merits of their inferences are fiercely disputed. This thesis investigates the dynamics of models of vocabulary evolution such as finite-state Markov chains, the Multistate model, and the Stochastic Dollo model. Using both simulated and empirical data sets, it evaluates their robustness and accuracy. The thesis identifies serious problems with the widespread use of finite-state Markov chains (such as the Covarion model), which can overfit linguistic data sets and infer incorrect language trees. It explores how models of vocabulary evolution can be improved to contribute more effectively to our understanding of linguistic prehistory.