Igor Yanovich is the project leader of an Emmy Noether junior research group hosted at the University of Tübingen. His work spans computational statistics, historical linguistics and theoretical linguistics, and includes collaborations with geneticists making use of mathematical evolutionary methods to perform inference about the linguistic and extra-linguistic past. Igor completed undergraduate studies at the Lomonosov State University and graduate studies at the Massachusetts Institute of Technology. He held postdoctoral positions at the University of Tübingen, including at the interdisciplinary Center for Advanced Study “Words, Bones, Genes, Tools”, and the Dept. of Philosophy at Carnegie Mellon University; his research stays included those at the Dept. of Translation and Language Sciences at the University Pompeu Fabra and the Dept. of Life Sciences and Biotechnology at the University of Ferrara, out of which the current project stems.
Computational Inference In Non-Tree Models of Language Diversification: The Case of Bantu
[Joint work with Silvia Ghirotto (Ferrara), Patricia Santos (Bordeaux) and Andrea Benazzo (Ferrara)]
In this talk, I will discuss three interrelated themes:
– models of language
-family differentiation: trees and their alternatives
– computational inference of linguistic (pre)history – the history of the Bantu language family
The Bantu languages are one of the largest language families of the world, spoken across vast areas in Sub-Saharan Africa, and featuring many hundreds of different languages. It is relatively well understood which Bantu languages are particularly closely related to each other – in other words, the shallow structure of the family is reasonably well-known. But the family’s deeper structure, and the early history of Bantu splits into major subgroups, remain elusive. This is arguably because the Bantu show a rather non-tree-like pattern of innovation spread. We will see how one can fruitfully think about such a linguistic history in terms of a “backbone tree” which is overlaid by the later copying processes through linguistic contact. With this conceptual model at hand, we can also perform computational inference to uncover statistically some aspects of the prehistory of the Bantu. I will informally introduce the technique of Approximate Bayesian Computation that makes inference in such complex historical models possible, show the just-in results our group has obtained, and explain the limits of inference that we discovered.