Using simulated learning to account for tone typology

Abstract

This study reports on the first investigation into the learnability of tone spread and shift patterns, as determined by computer simulations of the learning process. Our simulations are cast within the context of a synchronic analytical framework developed in earlier work. The framework uses licensing constraints and foot structure to drive various kinds of tonal reassociation. One problem with the framework was that it was able to generate various unattested patterns. We address this problem through the learnability simulations; our results show that the representable-but-unattested patterns are harder to learn, explaining their non-attestation. This way, we demonstrate that learnability simulations are a meaningful tool for a typological account of tonal phenomena.