Visual Similarity in Adult Word Extension and Artificial Word Learning

Reanalyzing adult word-extension data using DINOv2 visual embeddings reveals that distance to taught examples predicts extension, with a log odds decrease of 3.23 per standard deviation. Experimental results show that subordinate examples lead adults to extend words to only 1.1% of superordinate items, compared to 77.5% for superordinate examples.