Statistical models of sound change
Modeling probabilities of synchronic alternations based on diachronic factors
We develop a statistical model for deriving typology within the “historical bias” approach. Using a non-parametric bootstrap technique on surveys of sound changes, the model estimates the probabilities of synchronic alternations based on two diachronic factors: the number of sound changes an alternation requires and the respective probabilities of those sound changes. The model not only predicts that unnatural processes will be rare compared to natural processes, but also that some unnatural processes will be more frequent than others. For example, the model predicts that post-nasal devoicing is significantly more frequent than final voicing. Both predictions closely match the observed typology: post-nasal devoicing is attested as a synchronic alternation, whereas final voicing is not. This distribution is predicted neither under the synchronic approach to typology nor under alternative diachronic models.
