Children Actively Sample Information in Support of Learning New Words

• In Event: Active Learning Across Domains and Development: How Goals and Experience Shape Children’s Pursuit of Information

Abstract

Young word learners are faced with the difficult task of determining the meaning of thousands of words based on ambiguous input. How do children learn word meanings given limited and incomplete information? One way that children might confront this task is by intervening in their environment in ways that support learning. Throughout development, children are more than simply passive absorbers of information – they actively construct their own learning environment. This ability could have important consequences for how children simplify the problem of learning new word meanings. We present the results from two studies investigating how children seek out information during word learning and what consequences their sampling strategies have for learning.

In Experiment 1, we find that children (N=56; 3-8 years) sample words that reduce referential ambiguity about novel word meanings. In a cross-situational word learning experiment, we manipulated the referential ambiguity of label-object pairs experienced during training and investigated which words participants chose to subsequently sample additional information about. Two novel object-label associations always occurred with one another, such that it remained ambiguous which of the two objects each label referred to (ambiguous items; Figure 1A). Two other novel object-label associations occurred with a familiar object-label pair, allowing children to disambiguate reference based on mutual exclusivity (mutual exclusivity items). Children preferentially selected ambiguous items during a subsequent sampling phase (M=63.4% of trials, 95% CI=[54.4%, 72.4%], compared to a chance level of 50%; z=2.71, p=.007; Fig 1B). The tendency to select ambiguous items increased with age (z=2.48, p=.01; Fig 1C). These findings demonstrate that children will actively seek information to reduce referential uncertainty during word learning.

Does having active control over their input help children learn new words? In Experiment 2, we asked whether allowing children (aged 3-5 years) to actively control a portion of their training input during an explicit word learning task supports novel word learning. Participants (N=219) were exposed to a set of novel object-label associations that varied in exposure frequency. We then compared children’s word learning in an Active condition, in which they chose which words to learn about next, to two yoked passive conditions, in which participants viewed the same novel labels selected by a (yoked) participant from the Active condition. Children in yoked conditions had either identical (Yoked Passive) or different (Yoked Passive Exposure Mismatch) initial word learning experience compared to each (yoked) Active participant (Fig 2A). Our results support two main conclusions. First, children in the Active condition made informative choices, selecting words experienced less frequently during initial exposure (M=60.7%, 95% CI=[56.7%, 64.7%], t(72)=5.33, p<.001; Fig 2B). Second, actively selecting words that complement past learning experience matters. Children in the Yoked Passive Exposure Mismatch condition (M=77.1%, 95% CI=[72.1%, 82.0%]) - unlike children in the Yoked Passive Condition, M=81.3%, 95% CI=[76.8%, 85.8%]) - showed worse learning than children in the Active condition (M=84.8%, 95% CI=[80.6%, 88.9%], z=2.49, p=.01; Fig 2C).

Together, these studies shed light on the information-seeking strategies children use when learning new words and how these sampling strategies support word learning.

Authors

• Martin Zettersten, Princeton University
  Presenting Author

• Koeun Choi, Virginia Polytechnic Institute and State University
  Non-Presenting Author

• Heather Kirkorian, University of Wisconsin - Madison
  Non-Presenting Author

• Jenny R Saffran, University of Wisconsin - Madison
  Non-Presenting Author