Investigating the features of child-directed versus overheard speech in all-day recordings of Spanish-speaking families
International Congress of Infant Studies Virtual Congress
Recent studies of day-long recordings have shown that children's language environments consist of many different opportunities for caregivers to support their children's learning. Importantly, verbal engagement during child-directed speech, but not overheard speech, has been linked to children's later outcomes (Weisleder & Fernald, 2013). A growing list of verbal features has been hypothesized to explain why child-directed speech supports learning, including total number of words, number of different words, and number of conversational turns (Hart & Risley, 1995; Romeo et al., 2017; Rowe, 2012). However, few studies have compared how these features vary within the same families in the different settings in which child-directed and overheard speech occur (Foushee et al., 2013). Using data from Weisleder & Fernald, we examined three automated metrics from LENA (adult word counts (AWC), conversational turns (CTC), and child vocalizations (CVC)), exploring the extent to which these metrics, individually or in combination, predicted classification of child-directed versus overheard speech. Investigating these differences can help us better understand why childdirected speech, in particular, supports children's language learning. Participants were 29 primarily lower-SES Spanish-speaking families with 24-month-old children who recorded one typical day using LENA. Native Spanish-speaking coders listened to each 5-min segment of the all-day recording and classified each segment as: (1) primarily child-directed, (2) primarily overheard, or (3) 50% child-directed and 50% overheard (split). We conducted hierarchical logistic mixed models to examine the degree to which LENA-provided metrics predicted the classifications. Segments of split speech were divided equally into child-directed and overheard speech categories. All metrics were converted to rates/minute and mean-centered within each family to reduce collinearity and to allow interpretation of values as relative to each family's mean rates. In our first model, we tested whether AWC, CTC, and CVC each independently contributed to the probability of child-directed versus overheard speech. We found that AWC (B=-.44, 95%CI=[-.54,-.33]), CTC (B=.37, 95%CI=[.23, .52]), and CVC (B=.30, 95%CI=[.18, .42]) each contributed unique variance. Interestingly, lower AWC rates were associated with a higher probability of child-directed speech, whereas higher rates of CTC and CVC were associated with a higher probability of child-directed speech (Figure 1). In our second model, we tested whether the interaction between AWC and CTC contributed unique variance (Figure 2). That is, does CTC help our prediction of child-directed speech at all levels of AWC, beyond the unique and independent contributions for each metric? We found that while AWC (B=-.53, 95%CI=[-.63, -.43]) and CTC (B=.62, 95%CI=[.50, .74]) each still contributed unique variance, the interaction of both metrics was not significant (B=-.02, 95% CI=[-.09, .05]). These findings demonstrate that child-directed speech may, in part, be defined by lower rates of adult words and by higher rates of conversational turns and child vocalizations than in overheard speech. Ongoing analyses employ ROC curves to test the sensitivity and specificity of these metrics. This work is a critical step towards understanding what defines child-directed speech in naturalistic settings and elucidating how child-directed speech supports children's language learning.
Child and Adolescent Development
Janet Bang, Adriana Weisleder, George Kachergis, Virginia Marchman, and Anne Fernald. "Investigating the features of child-directed versus overheard speech in all-day recordings of Spanish-speaking families" International Congress of Infant Studies Virtual Congress (2020).