Phonological neighborhood density and intra-speaker variation in vowel production

Abstract

Words’ phonetic characteristics are affected by numerous lexical factors. Wright (2004) showed that words’ phonological neighborhood density (PND) affects the production of vowels. Vowels in words with high PND are hyperarticulated relative to the same vowels in low-PND words. Wright interpreted this as evidence that speech production accommodates listeners’ presumed needs: high-PND words are hyper-articulated because they are generally harder to perceive than low-PND words (Vitevitch & Luce, 1998). This effect has been the subject of numerous recent studies, including ones by Munson and Solomon (2004) and Munson (2013) that replicated the effect, and ones by Buz and Jaeger (2016) and Gahl, Yao, and Johnson (2012) that did not. One possible explanation for Wright’s original effect is that the ease with which low-density words are perceived allows speakers more freedom to produce the sounds in them variably, including under-articulating them. If this is true, then we would predict that the acoustic characteristics of low-PND words would show more trial-to-trial variability than those of high-PND words. To examine this, we recorded 19 speakers saying 10 repetitions each of high- and low-PND CVC words. The F1 and F2 frequencies of the vowels were measured. For each token, the distance from the center of each speaker’s F1/F2 space was computed. The coefficient of variation for this measure was higher (i.e., more variable) for low-PND words than for high-PND words, t[18]=−2.80, p = 0.011.