Looking Under the Hood of the Infraphonological Vocal Complexity Score

Background:

The infraphonological vocal complexity (IVC) score is derived from computer-automated analysis of day-long samples of children’s vocalizations according to 12 acoustic properties intended to describe the degree to which vocalizations are speech-like (Oller et al., 2010). The original IVC is the sum of these 12 properties weighted by regression coefficients from a multiple regression model that predicted concurrent chronological age of children who were typically developing (TD). Because we wish to derive a measure of the degree to which vocalizations are speech-like, IVCs with regression weights based on predicting productive language might be more valid than the original IVC. Because a valid measure of speech-likeness of vocalizations might have clinical value in minimally-verbal children with ASD, we need to test the validity of such measures in minimally-verbal children with ASD.

Objectives:

We sought to identify (a) the set of acoustic properties of child vocalizations with zero-order correlations with productive language at or above a moderate effect size (i.e., language-related properties), (b) the participant sample providing the most useful weights (i.e., typically developing, TD, or autism spectrum disorder, ASD), and (c) the relative predictive validity for later productive language in a separate sample of initially minimally-verbal children with ASD.

Methods:

The model building samples were from the LENA Research Foundation (LENA TD, n = 30; LENA ASD, n = 43) and had mean expressive language ages of 28 months (SD = 6) and 22 months (SD = 9), respectively. The productive language measure in the LENA samples was an aggregate of the Language Development Survey raw score and the Child Development Inventory Expressive age equivalent. The test sample of preschoolers with ASD (n = 20) had a median of 4 and mode of 0 words said according to parents when day-long vocal samples were recorded.  In the test sample, the interval between vocal samples and the productive vocabulary measure (i.e., MacArthur-Bates Words and Gestures checklist) was 4 months.

Results:

Results indicate that 8/12 and 9/12 properties had at least medium-sized associations with concurrent measures of productive language in the LENA ASD and LENA TD samples, respectively. Different properties predicted productive language in the two groups (e.g., Squeals in the ASD, Growls in the TD). In the test sample, the IVC variant based only on the language-related acoustic properties in the LENA ASD sample was the best predictor of later productive vocabulary (r = .54), relative to that based on language-related properties in LENA TD sample (r = .37).

Conclusions:

In the minimally-verbal sample of children with ASD, the size of the association between the selected variant of IVC and later productive vocabulary was  similar in magnitude relative to that for the original IVC (r = .51) (Woynaroski, 2014). Future research with a larger ASD sample to derive more precise regression weights is needed to determine whether the selected variant on the IVC improves the association with productive language over the original IVC.