Proceedings of the Annual Meeting of the Cognitive Science Society

One powerful way children can learn word meanings is viacross-situational learning, the ability to discern consistentword-referent mappings from a series of ambiguous scenes andutterances. Various computational accounts of word learninghave been proposed, with mechanisms ranging from storingand testing a single hypothesized referent for each word, totracking multiple graded associations and selectively strength-ening some of them. Nearly all word learning models as-sume storage of some feasible word-referent mappings fromeach situation, resulting in a degree of learning proportionalto the number of co-occurrences. While these accumulativemodels would generally predict that incorrect co-occurrenceswould slow learning, recent empirical work suggests these ac-counts are incomplete: paradoxically, giving learners incorrectmappings early in training was found to boost performance(Fitneva & Christiansen, 2015). We test this finding’s general-ity in a new experiment with more items, consider system- anditem-level explanations, and find that a model with error-drivenlearning best accounts for this benefit of initially-inaccuratepairings.