Calculating false cognates

Abstract

This paper presents an extension of Baxter & Manaster-Ramer’s (2000) approach to the problem of false cognates in the determination of relationships between languages. Their approach uses a Monte Carlo simulation to estimate how many lexical similarities we can expect to be due to chance between two lexical lists from different languages, and consequently how many are too many to be all false cognates. Although very efficient, their model has the shortcoming of being applicable only to simple lexical lists such as the Swadesh list, with one-to-one semantic correspondences between the individual terms. Here I present a new model that can be applied to any kind of word list, and can include comparisons between multiple terms sharing the same semantic field. After a theoretical description, a controlled test and a contra-test, I finally apply the method to a real test case, investigating the probability of relation between Pre-Greek, the nonIndo-European substrate of classical Greek, and Proto-Basque, Proto-Uralic and ‘Proto-Altaic’.