Introduction

The YorkTalk laboratory phonology speech generation system (Coleman 1992a, 1992b; Local 1992) implements two radical hypotheses about phonology and phonetics: (i) There are no segments in phonological or phonetic representations; (ii) There is no need for rewriting rules in the expression of phonological relations, regularities and generalizations, or in the phonetic interpretation of phonological representations. Instead, phonological representations are hierarchically structured graphical objects, and phonological relations, regularities and generalizations, and phonetic interpretation, are represented and solved by a declarative system of simultaneous constraints. Previous descriptions of the YorkTalk system have concentrated in particular on three areas: (i) The phonological structure and phonetic interpretation of single syllables in English (Coleman 1992a); (ii) The representation and interpretation of assimilation in English (Local 1992); (iii) Declarative analyses of apparently procedural phenomena, such as epenthesis, metathesis, and elision (Coleman 1990, 1992b).

In this paper, I shall describe the extension of the system from monosyllabic words (and assimilation between monosyllabic words) to polysyllabic words, with special attention to phonetic variability contingent on foot structure and the position of syllables within feet. Variability of this kind includes a diverse set of reduction phenomena, including vowel-quality reduction, vowel devoicing, vowel elision, flapping, and syllabic sonorants. All of these phenomena are modeled without rewrite rules in an explanatory, unified way within the YorkTalk system, and the quality of the resulting synthetic speech is both highly intelligible and natural.

The structure of this paper is as follows. In the second section, I shall briefly describe the theory of segmental phonology which underlies most current text-to-speech systems.