An experimental study comparing linguistic phylogenetic reconstruction methods

Abstract

This paper reports a simulation study comparing and evaluating the performance of different linguistic phylogeny reconstruction methods on model datasets for which the true trees are known. UPGMA performed least well, then (in ascending order) neighbor joining, the method of Gray & Atkinson and finally maximum parsimony. Weighting characters greatly improves the accuracy of maximum parsimony and maximum compatibility if the characters with high weights exhibit low homoplasy.