Preschoolers retrospectively re-evaluate word meanings upon learning that their source was unreliable

• In Event: 2-140 - Social mechanisms of selective trust and selective learning in childhood: New evidence and implications

Integrative Statement

Children use speakers’ past accuracy to make inferences about novel word meanings those individuals provide in the future (Koenig & Harris, 2005; Harris, 2012). This research typically establishes speakers as accurate or inaccurate and then asks children to learn novel words. An open question is whether children can retrospectively re-evaluate information based on learning that the source was inaccurate. One study that investigated this ability found that preschoolers can revise novel object-label mappings upon discovering that the speaker who introduced them was inaccurate (Scofield & Behrend, 2008). However, in this study, only 10 of the 24 tested children demonstrated this behavior. Thus, it remains a question whether preschoolers can make retrospective judgments and what mechanisms support this ability.
We addressed this question in three experiments. In Experiment 1, we tested whether 4-6-year-olds (N=48) can make retrospective reliability inferences. Children first heard a speaker introduce novel labels for novel objects, and then saw the speaker label familiar objects accurately or inaccurately. Children were then given retention and disambiguation trials. Children retained novel words learned from an accurate speaker better than from an inaccurate speaker (78% and 55% correct, respectively; Figure 1), Wald χ2(1, N=48) = 8.39, p = .004. Overall, however, children were better when asked to make a disambiguation inference than to retain the speaker’s original label.
In Experiment 2, we replicated the difference between retention and disambiguation, examining both a retrospective (backwards) and prospective (forwards) inference with a new group of 96 children. We observed a significant effect of the trial type by direction of inference interaction, Wald χ2(1, N=96) = 32.64, p < .001 (Figure 2). Children’s inferior performance on retention trials was related to the retrospective nature of reliability inferences and may have been linked to children’s developing memory capacity.
In Experiment 3 (N=26), we tested whether the memory demands associated with processing incorrect labels explain children’s inferior learning from inaccurate speakers. Speaker 1 performed a novel label training and then Speaker 2 generated accurate or inaccurate labels of familiar objects. There was no effect of condition on children’s word learning from Speaker 1 based on the accuracy of Speaker 2, suggesting that the effect of condition in Experiment 1 was unlikely to reflect the memory demands associated with processing inaccurate labels.
These results indicate that preschoolers continue to monitor the reliability of their sources of word knowledge after learning new words and update their beliefs as more reliability data becomes available. This observation is consistent with the “wait and see” strategy discussed in research on mutual exclusivity (e.g., Savage & Au, 1996): children keep in mind possible labels for an object for a period of time, until they receive more information that allows them to revise their hypotheses. Taken together, our findings extend Scofield and Behrend’s (2008) findings and present evidence that 4-6-year-olds can make retrospective reliability inferences and revise their beliefs about earlier learned word meanings accordingly.

Authors

• Elena Luchkina, Brown University
  Presenting Author

• Kathleen Helen Corriveau, Boston University
  Non-Presenting Author

• David Sobel, Brown University
  Non-Presenting Author