Rationale for a 'Many Maps' Phonology Machine

Abstract : M3P, our 'many maps' model of phonology, raises a number of questions about the nature of linguistic explanations and the ways in which connectionist models can contribute to the advancement of phonological theory. In this paper we attempt to answer some of the questions we and others have raised as a result of this work. For a variety of reasons, we view phonology as an attractive starting point for cognitive scientists seeking to understand language. The domain, sequences of phonemes, is purely symbolic. The operations are familiar: chiefly insertion, deletion, and mutation of elements. The structures involved are quasi-linear. (Some theories employ limited-depth trees to represent syllables or feature hierarchies, but phonology does not admit self-similar embedded structures or objects of unbounded depth of the sort required by syntax). True phonological processes are highly regular: they do not suffer from the plethora of special cases that complicates syntax and morphology.