Quantitative Analysis and Characterization of Natural Language Evolution Datasets

Methods for phylogenetic inference have been developed mainly for the reconstruction of evolutionary relationships of species based on biological sequence data. However, these methods are also made use of in linguistics for inferring phylogenies concerning the evolution of natural languages. In the scope of this thesis, we examine the corresponding linguistic input data. We conduct a case study on an exemplary morphosyntactic data set, examining various methods to analyze the signal it contains and to eliminate geographical information the data may include. Further, we perform analyses on numerous linguistic data sets collected from various sources and assembled in a database. We compare these data sets to morphological data from biology, considering differences in the behavior of phylogenetic inferences with RAxML-NG. Additionally, we investigate how it impacts the tree inferences, whether we represent a data set by a binary or by a multi-valued MSA. We study how to model subjectivity related with synonym selection in cognate data. We present probabilistic MSAs as a possible solution and show on an example data set that this might be an appropriate approach