Do Blind Spots Matter for Word-Referent Mapping? A Computational Study with Infant Egocentric Video