The phonetic interpretation of headed phonological structures containing overlapping constituents

Abstract

In this paper I shall present a theory of phonetic interpretation of headed phonological representations. The phonological representations in question are non-segmental, hierarchical, graphical objects similar to those in common use in autosegmental, metrical, dependency and ‘government and charm’ phonology, although the details of the phonological formalism I employ are different in some respects from each of these. The theory of phonetic interpretation is based on a parametric, dynamic model of phonetic representation. The distinction between ‘head’ and ‘non-head’ constituents is central to the phonetic interpretation model. As well as being formally explicit, I have developed a computational implementation of this theory, constituting a novel speech synthesis program, ‘YorkTalk’. Consequently, although the theory, like any other, is likely to contain certain faults, it goes beyond pencil-and-paper phonological theories, in that it is capable of algorithmically generating quite natural-sounding speech-like signals of a superior quality to other methods of generating synthetic speech, albeit only for isolated words.