Language trees, zipping and error estimation

A method was recently proposed to estimate distances between a pair of given texts. The distance estimation appeared to be reliable enough to infer a phylogenic tree of languages, even though no error estimation has been provided. This essay reviews the method and explains its application for inferring phylogeny on a collection of heterogeneous texts. An approach for estimating the confidence of the classification is introduced and the results are discussed. 1 Background and method A simple method for estimating a “distance” between a pair of given text was recently described in a paper [1], which attracted attention and criticism [2]. Its simplicity and the lack of assumptions on the nature of texts, however, make it an appealing technique for evaluating distance matrices. A (symmetric) distance matrix contains an entry for each pair of items, i.e., texts from different sources. The authors propose a distance DAB reported in the appendix and show ∗Questa è la tesina conclusiva del primo anno della scuola SAFI, scritta per il corso “Linguaggio ed Evoluzione” tenuto dai proff. Cavalli-Sforza e Wang.