In defense of discrete features

Abstract

A major challenge for discrete feature theories of speech perception is that the acoustic signal is continuously changing within and between segments at a high rate. These theories must explain how continuously changing spectral patterns are perceived as discrete and independent features. This problem does not disappear after auditory transformations of the acoustic signal, since temporal and frequency resolution, despite being coarse, still preserve most of the signal dynamics. In this work, it is shown that if articulatory features are extracted from the continuously changing acoustic signal, assuming that the signal is the output of a lossless vocal tract terminated in a unit resistance, i.e., a Darlington configuration (Iskarous, Journal of Phonetics, 2010), it is possible to obtain a discretely changing place of articulation feature from the continuous signal change. This will be illustrated using articulatory inversion of VV and CV transitions from the X-ray microbeam database for five male and five female participants. This work supports theories of discrete and independent perceptual features, as assumed for instance by Miller and Nicely (1955), if the acoustic signal is interpreted in terms of the vocal tract actions that cause it (Goldstein and Fowler, 2003).