Preliminary Report: Predicting Expressive Language Abilities at Age 2 from Fine-Grained Infant Vocalization Codes at 12 Months
Objectives: Determine whether vocalization coding adds significant explanatory variance to models of language ability from 12 to 24 months, in a sample of children at high- and low-risk of developing ASD.
Methods: Thirty-six children were assessed at 12 and 24 months as part of a longitudinal study of brain development (IBIS; Estes et al., 2015). The first group had an older sibling with ASD and were ultimately diagnosed with ASD (HR-ASD, N=10, 1 female). The second group had an older sibling with ASD but did not receive an ASD diagnosis (HR-neg, N=14, 6 female). The third group had a typically developing older sibling and were not diagnosed with ASD (LR-, N=12, 3 female). At each time point, participants were administered the Communication and Symbolic Behavior Scales (CSBS; Wetherby & Prizant, 2002), and Mullen Scales of Early Learning (MSEL; Mullen, 1995). Reliable annotators blind to risk status segmented and coded videotaped administrations of the CSBS at 12 months, and categorized infant vocalizations as speech- or speech-like (e.g., babbles, words) or non-speech (e.g., growls, squeals). The relative duration of speech- vs. non-speech vocalizations was explored as a potential predictor of language ability at 24 months (MSEL Expressive Language t-scores).
Results: After controlling for 12-month MSEL Expressive Language t-scores, the relative duration of speech- or speech-like vocalizations (vs. non-speech vocalizations) during the CSBS at 12 months accounted for significant additional variance (ΔR2) in 24-month Expressive Language ability, ΔF(1,33)=4.51, p=.04, β=.30. Overall, the combined model accounted for 34% of the variance in 24 month scores, F(2,24)=18.59, p<.001. Adding information about 12-month CSBS scores (total scores and standard scores for speech, communication, and words subdomains) did not significantly improve model fit, ΔF(4,28)=.59, p=.67.
Conclusions: The results of this preliminary exploration suggest that fine-grained vocalization coding in a heterogeneous sample adds additional explanatory variance to language predictions from 1 to 2 years of age, above and beyond what can be learned from standardized assessments. Our next steps are to increase our sample size, and include an additional time point with vocalization coding (6 months) to test the power of change in speech-like vocalizations from 6 to 12 months to predict 24-month language ability within separate diagnostic/risk groups.