Similarity structures among vowels

Abstract

Phoneticians have tried to predict phonological patterns by applying metalinguistic principles to phonetic substance. For instance, Liljencrants and Lindblom [Language (Dec. 1972)] assumed that vowel phonemes will tend to settle in maximally distinct locations in an acoustic vowel space. This presumes a knowledge of the acoustic-phonetic nature of vowel similarity. However, representations of the auditory-phonetic structure of a vowel space have usually been based on data about a stimulus set consisting of vowel sounds which approximate the phonemic norms of the language spoken by the experimental subjects. The results from these studies reflect more than the auditory qualities of the sounds, so they cannot be used without circularity for establishing phonetic explanations of phonological patterns. In the present study, 19 vowels were chosen to sample all areas of an “F1 by F2 by F3” acoustic vowel space appropriate for an adult male vocal tract. An American and a Persian subject judged the similarity of selected pairs of these sounds. Six dimensions (F1, F2, F3, F2−F1, F3−F2, F3−F1) were tested for independence and interdimensional additivity, using the ordinal properties of the subject responses. The data reject the possibility of expressing phonetic similarity over the whole vowel space as a simple combination of monotonic scales along any two or three of these dimensions. The similarity judgments among certain subsets of the stimuli, however, allow construction of simple expressions of similarity within subregions of the vowel space. [Supported by USPHS Grant 1-TO-HD00275.]