[PS-2.28] Statistical structure in speech to typically and non-typically developing toddlers

Onnis, L. 1 , Truzzi, A. 2 , Venuti, P. 2 , Esposito, G. 1, 2 & Edelman, S. 3

1 Nanyang Technological University
2 University of Trento
3 Cornell University

Parents naturally speak utterances containing partial self-repetitions (e.g., Want to get your ball? Get your ball? Do you want to get your ball?). Such 'statistically structured variation' contains cross-utterance statistical cues to the building blocks of language, and is predictive of children's lexical and grammatical structures. Here we compared structural characteristics of maternal language directed to toddlers with Autistic Spectrum Disorders (ASD, n = 25), Down Syndrome (DS, n=25), and a control group of Typically Developing toddlers (TD, n=31). We analysed the child-directed transcriptions of child-mother interactions during naturalistic dyadic play interactions. While children?s mean developmental age (24.60 months, SD=8.31) was the same across the three groups, we found that the proportion of sentences in sets of structured variation in child-directed speech was significantly the largest for ASD (55.3±2.8 %), smaller for DS (44.9±2.2%), and the smallest for the TD group (34.7±2.3%). Because statistically structured variation decreases with chronological age of the child, this finding is even more striking given that ASD, and DS children were on average 2 years (DS) and 3 years older (ASD). Our findings raise new questions on how parental speech shapes language development, and help clarify the link between statistical and language learning.