Unpacking the dynamics of cross-situational word learning strategies through trial-by-trial analysis: effects of word...

Word learning in natural environments often relies on cross-situational word learning (CSWL), whereby learners acquire word meanings from word-referent co-occurrences across different situations. Natural language follows Zipfian distributions where a few words occur very frequently while most occur rarely. However, how Zipfian distributions affect CSWL and how learning mechanisms vary across different contexts remain unclear. We addressed these questions in four CSWL experiments with adults. Experiments 1 (N = 36) and 2 (N = 38) found that Zipfian distributions may promote CSWL performance through prospective mutual exclusivity, whereby established mappings support the incremental disambiguation of new words, suggesting that the effectiveness of mutual exclusivity depends on contextual knowledge. Experiments 3 (N = 32) and 4 (N = 42) used trial-by-trial analyses to examine learning mechanisms under Zipfian and uniform distributions across varying levels of within-trial situational ambiguity, randomization, and contextual diversity. The Zipfian advantage emerged under high within-trial ambiguity, while its facilitative effect on associative learning required high contextual diversity. Moreover, hypothesis testing reduced reliance on associative learning under low within-trial ambiguity but enhanced it under high ambiguity. Together, these findings informed the Contextually Modulated Integrative Model of CSWL, in which the effects of Zipfian distributions and the interplay among learning mechanisms depend on learning context. The present study further advances the integrative model of CSWL and clarifies the sources of the Zipfian advantage.