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.