Prior knowledge bootstraps cross-situational learning

Recent research has highlighted the ability of adults as well as infants to learn word-to-world mappings over the course of statistical occurrences containing within-trial ambiguity. Participants in these studies are unfamiliar with all words and referents at the onset of these experiments, which differs from real-life word learning experience, in which learners are unlikely to enter a situation without any prior knowledge. Here we present a variation of the paradigm from Yu and Smith (2007) in which participants are taught a subset of the to-be-learned vocabulary prior to being exposed to the same training and test phases as in the initial experiment. Comparing results in conditions where participants had knowledge of some word-to-referent pairs prior to training with the data without prior knowledge shows the dramatic effects of prior knowledge in statistical word learning. Moreover, the behavioral data in this study provides valuable modeling constraints, as are demonstrated through simulations using several associative models of learning strategy