Bilingual statistical word learning in children and adults

• In Event: Lexical Development: environment and mechanisms

Abstract

Statistical learning, the ability to track regularities in the input, has been proposed to be key to language acquisition, including grammar learning (Gomez & Gerken, 1999), speech segmentation (Saffran, Aslin, & Newport, 1998), and word learning (Smith & Yu, 2008). In statistical word learning, children and adults can track the co-occurrences between words and referents over multiple learning situations to identify the correct meanings of words. Much of this research, however, has focused on the acquisition of statistics resembling a monolingual language environment, where learners learn to link a single word with a meaning. Bilingual language learning, in contrast, requires learners to track overlapping statistics, where multiple words may refer to the same meaning. Although recent research has found that adults are successful at bilingual statistical word learning (Benitez, Yurovsky, & Smith, 2016; Poepsel & Weiss, 2016), to date, no study has examined bilingual statistical word learning in children and whether this ability changes with age. In a pre-registered experiment, we test how 3- to 7-year-old children track bilingual word-meaning statistics compared to adults.

Learners were presented with a statistical word learning task where the co-occurrences between words and referents signaled the corred word-meaning links. Critically, two between-subject conditions manipulated the number of words co-occurring with each object. In the Single condition, each object was paired with a single word. In the Double condition, each object was paired with two words. Training was blocked such that, the first half of training presented 4 word-object pairings (First labels), while in the second half of training, either a new set of pairings was presented (Single condition) or the same objects were presented with second labels (Second labels; Double condition). Adults and children completed the exact same procedure.

Consistent with previous research, adults (Figure 1) in the Single condition (N=40, M=0.79, SD=0.27) and in the Double condition (N=37, M=0.67, SD=0.26) were successful at learning (comparison to chance, 0.5; Single: t(39)=26.82, p<0.001; Double: t(36)=21.01, p<0.001), with performance in the Single condition being statistically higher than the Double condition (𝜒2(1)=11.53, p<0.001). Similarly, children (Figure 2) in the Single condition (N=54, M=0.59, SD=0.33) and in the Double condition (N=52, M=0.56, SD=0.28) were successful at learning (Single: t(53)=18.19, p<0.001; Double: t(51)=22.65, p<0.001), with a trend for higher accuracy in the Single condition than the Double condition (𝜒2(1)=3.54, p=0.06). Comparison of adults and children revealed that First labels were learned better than Second labels (𝜒2(1)=10.24, p=0.001), an effect driven by the Single condition. Adults in general were much more accurate at statistical word learning than children (𝜒2(1)=18.51, p<0.001).

Our results provide the first evidence that children, like adults, succeed at bilingual statistical word learning. Further, similar to adults, children also trended toward preforming better at tracking a single word for a meaning than tracking two words for a meaning. Finally, adults performed much better overall than children, suggesting that statistical word learning improves with age. Our findings provide insights into statistical word learning from bilingual input, and highlight how statistical word learning changes across development.

Authors

• Viridiana L. Benitez, Arizona State University
  Presenting Author

• Ye Li, Arizona State University
  Non-Presenting Author

• Cassandra Leedom
  Non-Presenting Author

• Lia Ramalho Rocha, Arizona State University
  Non-Presenting Author