Evolutionary dynamics of structural features

Structural features have the potential to push the time barrier, after which we cannot test hypotheses about relatedness of languages, back in time. However, we have to know the stability of structural features in order to be able to apply them for such purposes. In this thesis I describe the typological profile of the Transeurasian languages, which serve as a data sample for the analysis of stability, build a phylogenetic tree with these languages, measure the stability of structural features as phylogenetic signal and evolutionary rate, reconstruct ancestral states of structural features and apply an admixture model from population genetics to test the performance of phonological, morphological and syntactic features in assigning languages to their respective language families and to investigate the level of diffusion in these three feature sets. More than half of structural features appear to have a high phylogenetic signal and evolve at a slow rate. I compare the stability across functional categories, parts of speech and language levels and come to a conclusion that argument marking (flagging and indexing), derivation and valency are the most stable functional categories, pronouns and nouns the most stable parts of speech and phonology and morphology the most stable language levels. The admixture model as implemented in STRUCTURE is able to correctly identify Turkic, Mongolic and Tungusic language families at the levels of morphology and syntax, whereas Japonic and Koreanic languages are assigned to the same ancestry. We see the least amount of admixture at the level of morphology and the highest level of admixture in syntactic features. One of the most important insights is that morphological features carry the most genealogical information, and these features could be used in the future to test relationships above the language family level.