Description:

Title: "Quantifying phonetic variation in imitated speech: An argument for acoustic landmark analysis"

Abstract: Speech is known to be highly variable across speakers and situations, and listeners pay attention to some of this phonetic detail for the rich contextual information it carries. In this talk I ask how much variability is present in speech, and whether some components of speech are more or less susceptible to variation. I present an approach to quantifying phonetic variation developed in collaboration with Stefanie Shattuck-Hufnagel (MIT), which approaches the question from the dual perspectives of perception and production. We analyze serial imitations of a heard utterance, where the linguistic object to be produced is fixed syntactically, lexically and prosodically, and employ a novel method for quantifying phonetic variation using acoustic landmarks (Stevens 2002) as correlates of phonologically-contrastive manner features. Imitated utterances produced by ten native speakers of American English resulted in 3500+ consonant and vowel landmarks (LMs), which were labelled and compared both to the lexically-specified LMs, and to the LMs produced in the stimulus. Our findings demonstrate and quantify systematicity in phonetic variation as measured in terms of LMs. They also reveal that speakers exercise choice in phonetic implementation, deviating both from lexical targets and from the phonetic detail of the heard stimulus. These results hold promise for the use of imitated speech in the study of phonetic variation, and for the use of LMs (and by extension other feature cues) as a phonologically grounded measure of variation in speech production.