Analytic bias in coocurrence restrictions

Abstract

The representation, content and learning of phonotactic constraints has spurred a lot of recent research phonology. This work concerns the constraints involved in the representation of place-based coocurrence restrictions. A single OCP-[PLACE] constraint is argued to compete in in constraint learning with other place-based and feature-based coocurrence constraints. The key prediction of this constraint participating in learning is generalization of a coocurrence restriction on to a novel place. This prediction, along with others, are tested in a series of artificial language learning experiments. A modification of the existing Hayes and Wilson (2008) phonotactic learner is presented (HWgain). This modified model is more robust in inducing gradient constraints, induces more general constraints and significantly improves on regression fit with English well-formedness ratings