Auditory robustness and duration of vocalic cues

Abstract

Building on an idea of Stevens [ICPhS (1980)] and Stevens and Keyser [Language (1989)], Lang and Ohala (LO) [ICSLP (1996)] hypothesized that the most commonly used features to differentiate vowels in languages would be those few that can be detected in a relatively short time. Other features could be used to augment the vowel inventory at the expense of requiring more time for their perception. They found support for this idea in a perceptual study of end-truncated English vowels. We conducted a similar experiment on the perception of 18 contrasting vowels in Cantonese including monophthongs and diphthongs and both rounded and unrounded front vowels. The stimuli were vowel fragments starting at 30 ms after an initial [s] and end truncated into pink noise at 20-ms intervals thereafter. Twenty Cantonese heard randomized sets of these stimuli and attempted to associate each with the vowel in 18 different key words symbolized by Chinese characters. The resulting confusion matrices of the shorter fragments revealed not only the expected high confusions between monophthongs and diphthongs but also between rounded and unrounded front vowels. The latter contrast, which is indeed used less in language’s vowel inventories, must require more time to perceive, supporting the LO claim.