4 June 2019
Time: 2:00 - 4:00pm
Venue: ArtsOne G.31
Vowel production in American English infant-directed speech
Infants' speech perception becomes aligned with the vowel distinctions that occur in their native language by the age of six to eight months (Tsuji, 2013). This learning process has typically been described as the result of distributional learning (Maye, Werker & Gerken, 2002) over specifically enhanced acoustic input (Kuhl et al., 1997). I present an extensive analysis of a corpus of both infant and adult-directed speech (IDS, ADS) as evidence against two claims in this domain. Firstly, a comparative analysis of IDS and ADS indicates that the properties of IDS vowel production are not consistent with greater discriminability relative to ADS. This analysis applied nuanced measures of discriminability to multidimensional data for an exhaustive set of vowels rather than formant measures for the three point vowels (/i/, /a/, /u/). Though speakers produced vowels that had greater dispersion in acoustic space when interacting with infants than with adults, vowel production in IDS was also more variable than ADS and therefore hindered the identification of relevant vowel distinctions. Secondly, a series of statistical models indicate that vowel categories cannot be recovered solely by exploiting the statistical properties of the acoustic input. A series of clustering models which replicated the use of statistical mechanisms in infancy did not successfully identify a set of fifteen vowel categories. As these models collapsed relevant distinctions, they indicated that exposure to caregivers' speech should reduce an infant's sensitivity to native language vowel distinctions. These analyses therefore prompt a reconsideration of methods used in comparative analyses of IDS and ADS as well as the mechanisms that are available to the infant learner.