Learning Verbs From Events: Semantic Similarity Differentially Affects Generalization Across Toddlerhood

• In Event: Poster Session 11
  In Poster Session: PS 11 Topic - Language, Communication

Abstract

Language learners face a daunting challenge in learning how generalize from limited linguistic experience and knowledge to describe novel events. Verbs pose particular difficulty for generalization; many studies find that young children demonstrate difficulty extending verb meanings to new events (e.g., Kersten & Smith, 2002). However, children’s generalization performance can improve in some learning conditions, such as by learning verbs in a series of perceptually similar events (Childers et al., 2016; Haryu et al., 2011). While prior work has focused on perceptual similarity, no work has highlighted whether and how semantic (categorical) similarity facilitates verb generalization and how children’s categorical knowledge influences these effects. We explore this question by manipulating the semantic (dis)similarity of objects during a verb learning and generalization task in toddlers.

We tested 48 toddlers between 21-30 months of age (24 younger Mage = 22.1 months; 24 older Mage = 28.3 months). In each of three blocks, children were taught two verbs, one in a semantically similar learning condition (with taxonomically related patients like a monkey/cow/dog) and the other in a semantically dissimilar learning condition (with taxonomically unrelated patients like tree/blanket/balloon). Verb learning was assessed in an eye-tracked word-recognition task on near generalization novel objects that were members of the previously viewed semantically similar category or far generalization novel shapes that were not in the same category and were less plausibly real-life objects. In test trials, toddlers viewed pairs of action videos and one was named. We assessed verb learning by measuring log looks to the named target action relative distractor from 300-2000 ms post label onset. We also collected parental report of words children say in categories that appeared in the semantically similar condition.

Using mixed-effects linear regression, we found a significant three-way interaction between learning condition, generalization condition, and age group. Notably, older and younger toddlers differed in their generalization of verb meanings in the far generalization condition. Here, the younger group was facilitated by semantically dissimilar learning contexts, whereas the older group was facilitated by semantically similar learning contexts. Further, beyond group, individual differences in category knowledge and age predicted children’s performance on far generalization. Category knowledge positively affected recognition for verbs from the semantically similar learning condition only, while age differently affected recognition for verbs from both learning conditions. There were no group or learning condition differences in the near generalization condition.

Our findings demonstrate that the semantic structure of children’s experience and knowledge affects verb generalization strategies, consistent with work showing changes in word learning mediated by vocabulary structure (Borovsky et al., 2016). Our results also suggest that younger toddlers flexibly recognize semantically unselective verbs – like those learned in the semantically dissimilar learning condition – which contrasts with findings that earlier-learned verbs are produced in relatively restricted contexts (e.g., Tomasello, 1992). This contrast suggests new avenues of inquiry into differences between verb recognition and production across development and learning conditions.

Authors

• Justin Kueser, Purdue University, West Lafayette
  Presenting Author

• Ryan Peters, Purdue University
  Non-Presenting Author

• Arielle Borovsky, Purdue University, West Lafayette
  Non-Presenting Author