Tree Model - Computational Phylogenetics in Historical Linguistics - Perfect Phylogenies

Perfect Phylogenies

The purpose of phylogenetic software is to generate cladograms, a special kind of tree in which the links only bifurcate; that is, at any node in the same direction only two branches are offered. The input data is a set of characters that can be assigned states in different languages, such as present (1) or absent (0). A language therefore can be described by a unique coordinate set consisting of the state values for all of the characters considered. These coordinates can be like each other or less so. Languages that share the most states are most like each other.

The software massages all the states of all the characters of all the languages by one of several mathematical methods to accomplish a pairwise comparison of each language with all the rest. It then constructs a cladogram based on degrees of similarity; for example, hypothetical languages, a and b, which are closest only to each other, are assumed to have a common ancestor, a-b. The next closest language, c, is assumed to have a common ancestor with a-b, and so on. The result is a projected series of historical paths leading from the overall common ancestor (the root) to the languages (the leaves). Each path is unique. There are no links between paths. Every leaf and node have one and only one ancestor. All the states are accounted for by descent from other states. A cladogram that conforms to these requirements is a perfect phylogeny.

At first there seemed to be little consistency of results in trials varying the factors presumed to be relevant. A new cladogram resulted from any change, which suggested that the method was not capturing the underlying evolution of languages but only reflecting the extemporaneous judgements of the researchers. In order to find the factors that did bear on phylogeny the researchers needed to have some measure of the accuracy of their results; i.e., the results needed to be calibrated against known phylogenies. They ran the experiment using different assumptions looking for the ones that would produce the closest matches to the most secure Indo-European phylogenies. Those assumptions could be used on problem areas of the Indo-European phylogeny with greater confidence.

To obtain a reasonably valid phylogeny, the researchers found they needed to enter as input all three types of characters: phonological, lexical and morphological, which were all required to present a picture that was sufficiently detailed for calculation of phylogeny. Only qualitative characters produced meaningful results. Repeated states were too ambiguous to be correctly interpreted by the software; therefore characters that were subject to back formation and parallel development, which reverted a character to a prior state or adopted a state that evolved in another character, respectively, were screened from the input dataset.