Manipulating phonological boundaries using distributional cues

Abstract

This study investigates the role of phonetic variability in category formation. The category boundary between /n/ and /m/ was manipulated in English-speaking adults. In Experiment 1 (n = 24), categories were indicated by lexical (novel word-object pairings) and distributional (within-category variability, category mean) information. During pre- and post-tests, listeners categorized stimuli from a 10-step re-synthesized continuum from "nado" to "mado". During training participants heard a bimodal distribution of tokens with one mode spanning 7 steps (wide) and one spanning 3 (narrow). For half the participants the wide category was "nado" and for the other half it was "mado". During training, tokens from each category were paired with different objects (socks or balls). A repeated-measures ANOVA showed that after training, categorization shifted toward the wide distribution (F(1,22) = 10.259, p < .01) and more so for ambiguous steps (F(9,198) = 5.283, p < .001). In Experiment 2 (n = 12), auditory information was identical to Experiment 1, but lexical information about the categories was removed (all training tokens referred to the same object). Listeners again shifted categorization towards the wide distribution (F(1,10) = 7.686, p < .05), indicating that distributional information alone was sufficient to change categorization behaviour. © 2013 Acoustical Society of America.