Investigating Distributions of Talk During Day-Long Naturalistic Recordings in English- and Spanish-Speaking Children
Society for Research in Child Development 2021 Virtual Biennial Meeting
Numerous studies have shown that the quantity and quality of early caregiver-child interactions are associated with children’s later language skills. Improvements in recording technologies have enabled researchers to gather day-long audio-recordings of child-caregiver interactions, yielding rich moment-to-moment records of the natural, spontaneous language used by caregivers. Further, speech-recognition algorithms provide automated counts of all adult words (AWC) spoken near to the target child. In most studies, researchers have relied on average AWCs, collapsed over the duration of the recording. Although these averages have predictive value (Gilkerson et al., 2019), they gloss over potential sources of variability in how language is distributed over the day (e.g., frequency and duration of dense periods of talk) (Casillas et al., 2019). Here, we explore novel measures of distributions of caregiver talk over the day. We ask: (1) Do families differ in how talk is distributed over the day? Are some more consistent across the day with no dense periods, while others have one or more periods of dense talk? (2) Is variation in the frequency or duration of dense talk associated with children’s language outcomes, beyond AWC?
Participants (n=98) were children from (1) primarily English-speaking families (n=54) with LENA recordings collected at 16 and 18 months and (2) primarily Spanish-speaking families (n=44) collected at 18 and 25 months. For each family, automated estimates of AWC per 5 minutes were derived using LENA, and converted to AWC/hour to control for variations in recording lengths. To find periods of dense talk, we identified the highest 30% of AWC segments for each family, and then applied a k-means clustering algorithm to identify one or more consecutive 5 minute segments that were similar in AWC (i.e., clusters of talk). We derived the number of clusters per family, converting to clusters/hour to control for variation in recording lengths. Cluster size captures the mean duration of those dense periods of talk per family. Children’s vocabulary size was assessed using the MacArthur-Bates Communicative Development Inventories (CDI) in English (Fenson et al. 2007) or Spanish (Jackson-Maldonado et al., 2003). Children’s language processing efficiency was assessed in the Looking-While-Listening task (Fernald et al., 2006). Accuracy reflects the time fixating on the named picture.
Analyses of the English-speaking sample indicated that children heard about 1676 words/hour (SD=588.75; range=434- 3057). In addition, children experienced 13.96 clusters/hour (SD=10. 85, range=0-35) with a mean cluster size of 2.06 (SD=1.69, range=0-8). As shown in Figures 1 and 2, clusters/hour was significantly correlated with Accuracy and CDI score (p<. 01). Critically, clusters/hour accounted for an additional 15% of the variance in CDI scores, after controlling for AWC (p = 0.012).
These results demonstrate that in addition to overall amount of talk, capturing variation in the frequency of dense periods of talk over the day provides a more nuanced view of the patterns of talk to young children. Importantly, frequency of dense talk provides additional predictive power when accounting for children's outcomes. Ongoing analyses of Spanish-speaking families will investigate if similar trends can be replicated in families from a different cultural and linguistic group.
Child and Adolescent Development
Joel Ramirez, Janet Yougi Bang, Anne Fernald, and Virginia A. Marchman. "Investigating Distributions of Talk During Day-Long Naturalistic Recordings in English- and Spanish-Speaking Children" Society for Research in Child Development 2021 Virtual Biennial Meeting (2021).