Many languages show systematic phonological differences between different parts of speech. Numerous mechanisms for capturing these differences have been proposed in the literature, but so far no complete picture has emerged of how learners decide that such mechanisms are necessary based on the data of their particular language. In this paper, two examples are contrasted: a case from Spanish in which learners appear to encode class-by-class differences even for properties which are not observably different across the classes, and a case from English in which learners evidently fail to encode class-by-class differences, even though there are observable statistical differences between the classes. It is argued that this difference can be captured by a hierarchical model in which the decision to subdivide grammar by classes is made globally, encouraged by a prior bias for simpler types of grammars.
Proceedings of the 26th West Coast Conference on Formal Linguistics
edited by Charles B. Chang and Hannah J. Haynie
Table of contents