Proceedings of the Annual Meeting of the Cognitive Science Society

Statistical learning is an active process wherein information isactively selected from the learning environment. As currentinformation is integrated with existing knowledge, it shapesattention in subsequent learning, placing biases on which newinformation will be sampled. One statistical learning task thathas been studied recently is cross-situational word learning(CSL). In CSL, statistical learners are able to learn the cor-rect mappings between novel visual objects and spoken labelsafter watching sequences where the two are paired togetherin referentially ambiguous contexts. In the present paper, weuse a computational method called Tensor Component Analy-sis (TCA) to analyze real-time gaze data collected from a set ofCSL studies. We applied TCA to learners’ gaze data in orderto derive latent variables related to real-time statistical learningand to examine how selective attention is organized in time.Our method allows us to address two specific questions: a) thesimilarity in attention behavior across strong vs. weak learn-ers as well as across learned vs. not-learned items and b) howthe structure of attention relates to word learning. We mea-sured learners’ knowledge of label-object pairs at the end of atraining session, and show that their real-time gaze data can beused to predict item-level learning outcomes as well as decodepretrained item knowledge.