Developmental differences in the role of temporal structure in ambiguous word learning

• In Event: It’s about time: Integrating the multiple timescales of attention and memory in early word learning.

Abstract

Learning words often requires integrating information about word meanings across ambiguous learning events distributed in time. Naturalistic naming events exhibit a “bursty” temporal structure, with the same words repeated across consecutive conversational turns. How does the temporal structure of word learning events affect what learners encode? How do children and adults differ in their ability to exploit temporal regularities? Here, we asked how 4- to 7-year-old children’s and adults’ performance in ambiguous word learning is influenced by the temporal distribution of naming events [https://osf.io/2hmxr/?view_only=6432022c65bf42aa9afcb865bc9ae37f].

Participants completed an ambiguous word-learning task with individual trials presenting 2 words and 2 objects, with no clear link between the two. Across trials, each word co-occurred most often with a single target object. In three between-subjects conditions, temporal structure was manipulated by varying the number of trials between repeated naming events for a given object (Fig 1). The Unstructured condition presented 1-6 trials between naming events for an object (M = 3 trials), the Massed condition presented 0–2 trials (M = 0.75), and the Interleaved condition presented 2–4 trials (M = 3). Adults and children were tested using the same procedure.

Children in the Unstructured condition (N=30, Mage=5.73 years) did not perform above chance (0.5) [M = .56, 95% CI = [.48, .64], p = .21; logistic mixed-effects model]. In contrast, children in the Massed condition (N=40, Mage=5.79 years) and in the Interleaved condition (N=40, Mage=5.94 years) did succeed at learning [Massed: M = .67, 95% CI = [.60, .74], p < .001; Interleaved: M = .64, 95% CI = [.57, .70], p < .001] (Fig 2A). The effect of condition was not significant [Wald χ2(2) = 5.09, p = .08]. These results show that children succeed at both massed and interleaved structure with no strong effect of temporal structure.

In contrast, adults showed a strong effect of temporal structure (Fig 2B). Adults were not successful at learning in the Unstructured condition [N=30; M = .60, 95% CI = [.56, .65], p = .16], while they successfully learned in the Massed [N=30; M = .86, 95% CI = [.77, .95], p < .001] and Interleaved [N=30; M = .68, 95% CI = [.58, .77], p = .009] conditions. There was an effect of condition [χ2(2) = 23.32, p < .001]. Adults in the Massed condition performed better than adults in the Unstructured [χ2(2) = 22.63, p < .001] and Interleaved [χ2(2) = 13.04, p < .001] conditions, demonstrating that massed structure boosts adults’ ambiguous word learning, an effect not present in children [condition X age group interaction: χ2(2) = 10.61, p = .005].

Our results show that temporal structure differentially affects children’s and adults’ ambiguous word learning: adults’ accuracy is boosted by massed structure, while children’s accuracy is not. Currently, we are increasing the memory demands of the task to explore if working memory skills explain adults’ unique massed advantage. Our findings provide insights into the role of temporal structure in ambiguous word learning and highlight how ambiguous word learning changes across development.

Authors

• Viridiana L. Benitez, Arizona State University
  Presenting Author

• Martin Zettersten, Princeton University
  Non-Presenting Author

• Erica Wojcik, Skidmore College
  Non-Presenting Author