Contextualizing the Cross-Situational Word Learning Task

Abstract

A majority of the current research on cross-situational word learning involves the use of a simplified lab task to simulate the referential ambiguity in naturalistic word learning scenes. The task’s presentation of un-occluded objects that are equally spaced on white backgrounds is clearly lacking in the external validity of the naturalistic word learning scene. Research on the first-person view of an early word learner depicts the naturalistic word learning environment as cluttered and rich in contextual information. The present studies examine whether including an interrelated context, through the connection and orientation of objects in the cross-situational word learning task, will affect participant accuracy in learning, tracking and recalling these word-referent pairs in referentially ambiguous scenes. In Study 1 (preregistered), we tested 24 children (4-6), and 50 adults on their ability to learn six novel word-object pairs. Participants were presented with three word- referent pairs in the interrelated context format of the cross-situational word learning task, and three word-referent pairs in the baseline (no context) format typically used in cross-situational word learning studies. Their ability to identify targets and their label recall rates were compared across conditions using a multilevel analysis of performance. Child participants exhibited a significant advantage in target selection trials for objects presented in the interrelated context format. In Study 2, we broke down the spatial qualities of the interrelated context from Study 1 by replacing it with a contact condition in which objects in the experimental condition were only touching each other. Child accuracy rates were compared across contact and baseline (no contact) conditions and demonstrated no significant differences in performance between formats, however adults showed a marginal effect in the opposite direction, suggesting that proximity may hurt target selection. Overall the results suggest that the interrelated context boost may result from a more nuanced level of encoding. Keywords: referential ambiguity; word-referent learning; cross-situational word learning