Determining the Evolutionary History of Languages 1 Donald

1 Evolutionary trees The evolutionary history of a taxonomic group, both in biology and in linguistics, is usually represented as a tree. Many of the actual trees that one nds in the literature, both biological and linguistic, are controversial for a variety of reasons. Sometimes the evidence from which a tree could be constructed is meager or ambiguous; very often it is \noisy", so that a tree cannot be derived cleanly. But it is also true that the methodology by which evolutionary trees are constructed is not very satisfactory. In this paper we will show that a well-known set of controversies in historical linguistics can be resolved by a new tree-construction methodology combining ideas from biology, techniques from computer science, and scholarship in historical linguistics. A problem that has generated controversy in historical linguistics for more than a century is nding the correct evolutionary tree for the Indo-European (IE) family of languages. The existence and membership of the family are not in doubt; neither is the fact that there are ten subfamilies of IE that include languages still spoken or adequately recorded. (There seem to be a few more that include extinct languages very sparsely recorded, but we must leave them aside for sheer lack of evidence.) But if you look at a \traditional family tree" of IE, it will show all ten major subgroups evolving independently from the ancestor language, Proto-Indo-European (PIE; see g.1) and that cannot possibly be correct, if only because the simultaneous ten-way speciation of a single language community is almost unimaginable under neolithic conditions. What a tree like this really indicates is that specialists cannot determine the correct rst-order branching of the tree using traditional methods. In fact, while historical linguistics has long had a limited but highly reliable method for evaluating evolutionary trees, it has never had a rigorous method for constructing all the optimal evolutionary trees for a language family (by any measure of optimality). 2 Lessons from biology Given this methodological deeciency, it makes sense for linguists to ask how biologists evaluate and construct evolutionary trees. In biology taxa can be represented by biomolecular sequences or morphological features, such as the presence or absence of vertebre to describe their data sets. Both columns from alignments of biomolecular sequences and morphological properties can be encoded as qualitative characters, which describe taxa by partitioning the taxa into the distinct states of the character. Sometimes …