Distributed Multiple Alignment for Historical Linguistics

摘要

This paper describes a highly distributed project in Data Mining. The program adapts an algorithm originally developed for use in bioinformatics to the needs of Historical Linguistics to automate what historical linguists call the Comparative Method by which similarities between vocabulary items in different languages are accounted for by devising rules of sound change that explain their divergence from a common ancestor. The algorithm is computationally very expensive, but the problem of massively searching for vocabulary items with a high degree of similarity really consists of many data-independent problems and is thus and ideal application of distributed methods. The system is implemented using the Parallel Virtual Machine on a non-homogeneous collection of machines.