Correction to “Introducing meta-analysis in the evaluation of computational models of infant language development”

Abstract

Cruz Blandón, M. A., Cristia, A., Räsänen, O. (2023). Introducing meta-analysis in the evaluation of computational models of infant language development. Cognitive Science, 47(7), e13307. https://doi.org/10.1111/cogs.13307

On page 15, a citation to Bunce et al. (2021; pre-print) inaccurately attributes an estimate of 5.82 h of daily infant speech exposure to their study.

Bunce et al. (2021) did not directly report on infants’ daily speech exposure. Instead, our estimate of 5.82 h of speech per day was derived from their data as follows: we first calculated the average rates of target-child-directed speech (TCDS) and adult-directed speech (ADS) per hour across the five languages studied (Table 2 in Bunce et al., 2021). The sum of these average rates—3.72 min per hour for TCDS and 10.84 min per hour for ADS—was then multiplied by 24 h to estimate full-day exposure, yielding 5.82 h per day.

However, this estimate excludes speech directed at other children but heard by the target child, accounting for an additional 4.61 min per hour as reported in the supplementary material of Bunce et al. (2021). Additionally, the estimate assumes the long-form recordings analyzed are representative of a full 24-h day, likely overestimating language exposure by including nighttime, when infants and their caregivers are typically asleep. The long-form recordings analyzed by Bunce et al. (2021) and the actual language input to infants is likely biased toward the waking hours of adults and children in the language environments studied. The estimate of 2124 h of speech heard per year presented in our paper is thus on the upper end of the likely input scale but remains within plausible bounds. For context, Hart and Risley (1995) report 45 million words heard by the age of 4 in families of the professional class, equivalent to about 937.5 h of speech (assuming an average word duration of 0.3 s), but this estimate is only for child-directed speech (CDS). Bunce et al. (2021) found that infants exposed to North-American English hear twice as much ADS as CDS, and our simulations aimed to account for all speech a learner hears.