Parallel adjustment of phonetic targets in L2 English voice onset time

Abstract

The phonetic realization of any given speech sound varies considerably across speakers and languages. For instance, the voice onset time (VOT) of [kh] can range from ~40 ms to over 100 ms (Cho & Ladefoged, 1999). Within American English, recent research has demonstrated that the VOT of [kh] is also systematically and linearly related to the VOTs of [ph] and [th] (Chodroff & Wilson, 2017). In the present study, we investigated whether these relations were maintained across L2 speakers of English. L2 speech production could arise from independent acquisition (or adjustment) of the phonetic targets for each speech segment. Alternatively, the presence of covariation would indicate that properties of speech sounds may be altered in tandem. To investigate this, VOT data was obtained from a subset of the ALLSSTAR Corpus, which contained matched connected speech data from 26 L1 American English speakers and 114 L2 English speakers (22 unique L1s). Preliminary analysis revealed strong VOT covariation among aspirated stops across L2 speakers (rs = 0.67 to 0.74), and qualitatively equivalent covariation across L1 speakers (rs = 0.57 to 0.75). The observed covariation may arise from a universal principle of uniformity requiring near-identical implementation of the shared laryngeal feature value.The phonetic realization of any given speech sound varies considerably across speakers and languages. For instance, the voice onset time (VOT) of [kh] can range from ~40 ms to over 100 ms (Cho & Ladefoged, 1999). Within American English, recent research has demonstrated that the VOT of [kh] is also systematically and linearly related to the VOTs of [ph] and [th] (Chodroff & Wilson, 2017). In the present study, we investigated whether these relations were maintained across L2 speakers of English. L2 speech production could arise from independent acquisition (or adjustment) of the phonetic targets for each speech segment. Alternatively, the presence of covariation would indicate that properties of speech sounds may be altered in tandem. To investigate this, VOT data was obtained from a subset of the ALLSSTAR Corpus, which contained matched connected speech data from 26 L1 American English speakers and 114 L2 English speakers (22 unique L1s). Preliminary analysis revealed strong VOT covariation among aspirat...