The distribution of data in word lists and its impact on the subgrouping of languages

Abstract

This work reveals the reason for the bias in the separation levels computed for natural languages with only a small amount of residues; as opposed to stochastically normal distributed test cases like those presented in Holm (2007a). It is shown how these biased data can be correctly projected to true separation levels. The result is a partly new chain of separation for the main Indo-European branches that fits well to the grammatical facts, as well as to their geographical distribution. In particular it strongly demonstrates that the Anatolian languages did not part as first ones and thereby refutes the Indo-Hittite hypothesis.