Phonetic and Phonological Representation of Stop Consonant Voicing

Abstract

This paper argues for a more structured view of the relation between the phonological feature [voice] and its specific phonetic implementations. Under the theory of universal phonetics proposed here, the implementation of [voice] is sharply constrained: the opposition is defined relatively, as more or less voicing, along a dimension consisting of exactly three discrete, ordered categories, which can be shown to have clear articulatory and acoustic bases. While the phonological feature allows certain rule equivalences across languages to be expressed, the phonetic categories describe possible contrasts within languages, and express markedness relations.* 1. INTRODUCTION. It is common practice to represent sounds in different languages which are phonetically similar, but not absolutely identical, with the same symbol or set of feature specifications. Sometimes this practice is simply a convenience, for ease in transcription or typesetting. At other times, the sounds in question are said to be different only at the phonetic level; at the phonemic level, they may indeed have the same feature specifications, and so are thought of as the 'same' sound. For example, the distinction in the SPE model (Chomsky & Halle 1968) between systematic phonemic vs. phonetic levels makes such a view possible. However, I will argue here that the version of the model proposed in SPE can be improved, allowing a more adequate treatment of a variety of facts. I will discuss how surface phonetic variation, within and across languages, can be derived in a synchronic grammar from the interaction of three relatively simple systems: the possible phonological features and their values, their possible phonetic category mappings, and phonetic detail rules accounting for variation within these phonetic categories. More generally, this paper contributes to an -important goal of linguistic phonetics: that of relating discrete and timeless phonological units to physical reality, with its continuous articulatory and acoustic manifestations. 1.1. THE SPE MODEL represents lexical items as matrices of binary-valued phonetic features; each row is a feature, and each column a segment. Phonological rules may change the values of features, or may add or delete segments, but may not change the inventory of features which form the rows of the matrices. By contrast, phonetic rules convert the binary values into quantitative values along continuous phonetic scales; these rules specify the value along each scale at which the given language divides the scale into its phonetic * This research was supported in part by a postdoctoral fellowship at MIT from NIH, by a research grant from UCLA, and by an NSF grant to Peter Ladefoged. I would like to thank Elan