BY
1 November, 2011 (Tuesday)
4:30 p.m. - 6:00 p.m.
LE2 (Library Extension 2)
ALL ARE WELCOME!
Abstract
Natural speech is full of variation. If we examine a sample of spontaneous speech of one speaker, we will find that some words tend to be phonetically reduced (shortened, with more centralized vowels, etc), whereas other words tend to be lengthened, articulated with more dispersed vowels and higher articulatory target maintenance. Sources of such pronunciation variation have been intensely debated. “Listener-oriented” accounts attribute variation to speakers’ attempts to maximize intelligibility; “speaker-centric” accounts attribute variation to fluctuations in production difficulty. A difficulty in adjudicating between these two accounts is that most factors facilitating production also facilitate recognition. For example, high frequency, concreteness, and familiarity all promote phonetic reduction – and they all increase speed and accuracy of comprehension and production, other things being equal.
A lexical property that avoids this ambiguity is phonological neighborhood density, i.e. the number of words in the lexicon that are phonologically similar to a given word: High density is associated with short naming latencies and high accuracy of production (Vitevitch, 2002), but also with decreased speed and accuracy of recognition (Luce & Pisoni, 1998). Because of this seemingly paradoxical property, neighborhood density provides a means of investigating the sources of pronunciation variation: Speaker-centricity would predict phonetic reduction of high-density words, whereas listener-centricity would predict higher articulatory target maintenance.
We examined the effect of neighborhood density on both word duration and vowel dispersion in a corpus of spontaneous speech (the Buckeye corpus, Pitt et al., 2007), for all monomorphemic CVC content words containing monophthongs other than schwa (n = 11,832). We fitted two mixed-effect regression models, one on word duration and the other on degree of vowel dispersion, taking into account phonological neighborhood density as well as other factors that could cause variation in speech (e.g. vowel target, speaker sex, age, word frequency, phonotactic probability, part-of-speech, phonological context, contextual predictabilities, speech rate, presence of disfluencies, number of previous mentions, etc).
The models reveal that, other things being equal, words with many neighbors have shorter durations and slightly more centralized vowels, compared with words with fewer neighbors. Our findings provide support for the speaker-oriented hypothesis over the listener-oriented hypothesis, and suggest that word-level pronunciation variation is more heavily influenced by features of the speaker’s own production system than by the consideration of listeners’ needs. This work also has implications on models of lexical production.
About the Speaker
Dr. Yao Yao is currently an assistant professor at the Department of Chinese and Bilingual Studies at the Hong Kong Polytechnic University. Before joining HK PolyU, Dr. Yao got her Ph.D. degree in linguistics from UC Berkeley. Her karma with Hong Kong dated back to the pre-linguistics era, when she did her bachelor’s study in information technology at the City University of Hong Kong. After discovering linguistics, she has worked on various projects that aimed to discover the structure of language and its representation in mind, using both experimental and quantitative methods. Her current research interests include lexical processing and production models, experimental phonetics and corpus linguistics. The languages she works on the most are English and Mandarin Chinese.