Speaker: Gašper Beguš (Harvard)
Title: Learning the Blurring Process
Date and time: Monday, April 2nd, 5:00-6:30pm
Location: 32-D831
Abstract:
Many artificial grammar learning experiments have provided strong evidence for the assumption that learning biases influence phonological typology: typologically rare processes have been shown to be more difficult to learn. Most of the experiments, however, fail to control for diachronic influences: in many cases, the observed typology can be explained by diachronic factors equally well. In this talk, I present a method for controlling for diachronic influences when testing learning biases. Unnatural processes provide a crucial solution to this problem. I show that a statistical model of diachronic development (that I call Bootstrapping Sound Changes) identifies a crucial mismatch in predictions between the learning and diachronic bias approaches. This mismatch allows me to design experiments such that diachronic factors are controlled for. I present results from two experiments that test learnability of complex vs. unnatural processes and suggest that learnability differences have to influence observed typology in cases that cannot be explained by historical factors. I also discuss implications of this approach for phonological theory in general.