Rooting through Direction – New and Old Approaches

In this paper two new methods to find the root of an unrooted tree are being presented. Unrooted trees are the output of some programs generating genealogical hypotheses for historical linguistics, stemmatology and some other disciplines. By example of stemmatology, the science of establishing the copy history of ancient texts we implement 4 approaches (Haigh 1970, Marmerola et al. 2016 and two own approaches) and test them on 3 benchmark datasets. For direction detection we use psycholinguistically determined letter confusion probabilities (Geyer 1977, Paap 1982). If one could assign the correct direction to edges in an unrooted tree, one would be able to locate root, but error rates may be a problem. We present Highest Direction Change Conversion Point Rooting which instead uses leaf-leaf paths and thus more context. Results point to a successful applicability.