Disentangling acoustic measures from lexical statistics in child-directed speech

Abstract

Acoustically, North American child-directed speech is hyperarticulated: it is slower than adult-directed speech and the vowel space is more expanded. Lexically, child-directed speech is composed of shorter, more frequent words, from denser neighborhoods. The issue is that hyperarticulation, including vowel space expansion, varies according to word structure and statistics: speakers hyperarticulate low-frequency words and phonetically reduce words from denser neighborhoods (Bell et al., 2009; Gahl etal., 2012). Since the acoustic and lexical parameters of child-directed speech co-vary so tightly, it may be important to disentangle them when modeling the effects of child-directed speech for developmental outcomes. In this longitudinal study, we measured a battery of acoustic-lexical characteristics of child-directed speech from n = 84 caregivers speaking to their infants aged 7, 10, 18, and 24 months. We then fit a series of models to regress out how eachcharacteristic independently predicted the children’s vocabulary sizes and phonological processing at 24 months. Results confirmed that many characteristics, such as vowel space, coarticulation, and speaking rate, co-varied with the lexical statistics. We nevertheless founddirect effects of coarticulation, as well as word frequency, length, and diversity, on the children’s development, demonstrating the need to carefully disentangle these co-varying parameters going forward.