English-hearing toddlers learn words from consistent and variable labels.

Early word learning is traditionally characterized as a process of mapping a single label onto a single referent. Yet, across many languages, child-directed speech (CDS) frequently features multiple labels for the same referent, including diminutivized and reduplicated word forms (e.g., doggy or woof-woof for dog). We examined how consistent versus variable labels impact children's novel word learning. In a within-subjects eye-tracking experiment, we exposed 18- to 36-month-old English-hearing children to novel objects that were named with consistent labels, with typical CDS variants, or with atypical CDS variants. Using looking-time measures of word-recognition accuracy, we found that children reliably learned words in the consistent and typical variation conditions, with weaker performance in the atypical variation condition. Using pupil size synchrony measures of moment-by-moment attention, we found that engagement was higher in both variation conditions compared to the consistent condition, and was highest in the typical variation condition. Our findings suggest (1) that children can successfully contend with multiple types of label variation when learning new words, and (2) that increased attention is a plausible mechanism by which label variation could support-rather than hinder-children's early word learning.