At this week’s MCQLL meeting, Andrei Munteanu will be presenting “Evaluating comparative reconstructions probabilistically”. Abstract below.
We will be meeting this Tuesday January 23rd at 3:00PM. Meetings are held both in person in room 117 of the McGill Linguistics department and on zoom.
In this talk, I present Wordlist Distortion Theory, a framework for the probabilistic evaluation of comparative reconstructions in historical linguistics. The framework estimates the likelihood that a randomly generated wordlist merits the same type and number of diachronic transformations (e.g. sound changes, replacements, etc.) as required by the reconstruction. The framework is primarily intended as a platform for objective and accessible debate surrounding spurious reconstruction in historical linguistics. Additionally, it can be used as a tiebreaker between conflicting reconstructions for the same data. Finally, the framework allows for probability-based theoretical arguments in historical linguistics about the interaction of synchronic and diachronic factors with reconstruction reliability.
I also present results from a case study, where Wordlist Distortion Theory was incorporated into a machine learning algorithm. The algorithm suggests comparative reconstructions (i.e. series of sound changes) stochastically while giving preference to those that decrease the probability of a spurious match. When tested on wordlists from 74 Austronesian languages and 5 proto-languages, the algorithm yields reconstructions that appear in line with general knowledge about sound change and about Austronesian historical linguistics.