The Effect of Distributional Restrictions in Speech Perception: A Case Study From Korean and Taiwanese Southern Min

In Korean, voiced oral stops can occur intervocalically as allophones of their voiceless lenis counterparts; they can also occur initially as variants of nasal stops as a result of initial denasalization (e.g., /motu/→[bodu] "all"). However, neither [ŋ] nor [ɡ] (the denasalized variant of the velar nasal) is allowed in the initial position due to the phonotactic restriction against initial [ŋ] in Korean. Given the distribution of nasal and voiced stops in Korean, this study draws on the idea of cue informativeness, exploring (a) whether Korean listeners' attention to nasality and voicing cues is based on the distributional characteristics of nasal and voiced stops, and (b) whether their attention can be generalized across different places of articulation without such linguistic experience. In a forced-choice identification experiment, Korean listeners were more likely than Taiwanese listeners to perceive items on the voiced oral-to-nasal stop continua as nasal when they occurred in the initial position than in the intervocalic position, with the exception of velar stops. The results demonstrate that the Korean listeners attended to the nasality cue more reliably in the medial position than in the initial position, since the nasality cue in this position is less informative due to initial denasalization. Two additional forced-choice identification experiments suggested that upon hearing initial velar nasal [ŋ], Korean listeners variably employed different perceptual strategies (i.e., vowel insertion and place change) to repair the phonotactic illegality. These findings provide support for exemplar models of speech perception in which cue attention is specific to the position of a word, and to segments rather than to features.