Feature economy as an organizational principle of sound systems

Nick Clements

CNRS / Université de Paris 3

Why are speech sounds composed of features? The classical answer to this question was proposed by Martinet, who after signalling the economy achieved by representing morphemes in terms of a small number of phonemes, states "a new economy is achieved by making phonemes result from combinations of non-successive phonic features, which further reduces the number of basic elements" (Économie des changements phonétiques, 1955, p. 95). In Martinet's view, features serve to reduce the number of primitive articulations required to implement a phoneme system. The central claim of feature economy is that features typically combine with maximal information-theoretic efficiency in defining speech sounds, subject only to the external limits imposed by articulatory and perceptual constraints. Feature economy is thus quite different from the more familiar notion of parsimony (phoneme economy), an analytical criterion which favors smaller phoneme systems over larger ones. Feature economy has been little discussed in the recent literature in functional phonology (e.g. Flemming 1995, Padgett 2001), which has tended to concentrate instead on the notion of maximal contrast or dispersion (de Groot 1931, Martinet 1952, 1955, Liljencrants & Lindblom 1972, Lindblom 1986). An exception to this trend is Boersma (1998), who discusses feature economy under the heading of symmetry, without, however, relating it to Martinet's original proposal. Feature economy and dispersion are near-contradictory principles, since while feature economy predicts that sound systems tend to be self-organized around a small number of articulatory/acoustic parameters, dispersion theory predicts that sound systems will be maximally dispersed in auditory space and should therefore multiply feature categories in an uneconomical fashion. Dispersion has been applied most successfully to the study of vowel systems and secondary articulations, but has so far achieved little success in the study of consonant systems. As Boersma points out, dispersion-based theories fail to account for such fundamental aspects of sound systems as phoneme symmetry and feature organization, in vowel and consonant systems alike. While Martinet applied feature economy primarily to the study of sound change, it is equally and more easily testable in the domain of synchronic structure, where it makes precise predictions. This paper will show that feature economy outranks dispersion as a basic organizational principle of sound systems. Predictions of feature economy are tested against the data collected in the UPSID data base of 451 phoneme systems (Maddieson and Precoda 1989). Some of the further questions to be addressed, as time permits, include the following: Is feature economy restricted to phoneme systems, or does it apply to the sounds occurring in output representations as well? Are the features active in feature economy patterns better defined in terms of primitive articulations (such as bilabial, labiodental, etc.), as originally suggested by Martinet, or in terms of more general feature categories (such as labial)? Are such features uniquely articulatory, or can acoustic-perceptual features such as [grave] be validated as well? More generally, what types of features best account for feature economy effects?