Links between Language Input in Infancy and Vocabulary in Childhood

• In Event: The Ins and Outs of Language Development: Integrating Multiple Measures from Inside and Outside the Laboratory

Abstract

While all typically-developing children learn the common words in their language(s), no two children have identical learning trajectories or language input (even twins; Elo, 2016). This raises a critical question: how does language input relate to language learning? We tackle this question by examining links between previously identified “useful” properties of children’s early language input and the same children’s vocabulary at three time-points.
We measured input and vocabulary links in a longitudinal sample of 44 English-learning American children. Our language input measures are derived from monthly day-long audio recordings and hour-long video recordings of these children from 6-17 months. Recordings were annotated for concrete nouns directed to the infant. Input measures were aggregated across all recordings for each child (~80 hours annotated input/participant). Our language learning measure is vocabulary: CDI-production at 1;6 (Fenson, et al., 1993, log-transformed, n=44), QUILS-vocabulary at 3;6 (Golinkoff, et al., 2017, n=34), and PVT at 4;6 (Weintraub, et al., 2013, n=34).
We used 5 input predictors from previous research: number of word tokens, number of talkers, proportion of short utterances (<3 words, a proxy for isolated/edge words), proportion of imperatives, and proportion of object presence (e.g. Bergelson & Aslin, 2017; Brent & Siskind, 2001; Hoff & Naigles, 2002). Backward model selection by AIC was used to identify the best model for predicting vocabulary at each age from these measures (see Fig. 1 for zero-order Spearman correlations).
Vocabulary at 1;6 was best predicted by a model with just tokens (adjusted-R2=0.12, p=.011): more noun tokens in 6-17 month input resulted in higher vocabulary. At 3;6, the best model included tokens and proportion of imperatives (adjusted-R2=0.38, p < .001), such that more noun tokens and fewer imperative utterances resulted in higher vocabulary. However, only proportion of imperatives was a significant predictor on its own (𝛽imp=-176.05, p < .01;𝛽tokens=0.001, p=.16). Age 4;6 vocabulary was best predicted by only the number of noun tokens (adjusted-R2=0.15, p=.014). See Figure 2.
Summarily, we find evidence that the overall amount of speech children hear early in life (for which noun tokens are our proxy) is consistently linked to vocabulary from infancy into childhood. Intriguingly, while imperative utterances made up a small proportion of the input (M=6.9% in this sample), this metric also predicted vocabulary, but only at age 3, with more imperatives predicting lower vocabulary. Imperatives direct the child’s attention instead of following their interests (Tomasello & Todd, 1983), and are likely linked to larger constructs concerning parental scaffolding and parent-child interaction. Notably, while all considered input metrics have been found to link to language learning in prior work, very few of them parsimoniously predicted later vocabulary in this study. This in turn suggests that neither language input nor vocabulary are monolithic or intrinsically stable: while the quantity of input exerts a clear (and likely ongoing) influence on children’s vocabulary, other components of the input that undergird learning in infancy may not be readily linked to children’s vocabulary across the first five years of life.

Authors

• Shannon Dailey, Duke Univeristy
  Presenting Author

• Federica Bulgarelli, Duke University
  Non-Presenting Author

• Elika Bergelson, Duke Univeristy
  Non-Presenting Author