Acquiring multiple meanings for a word is often proposed to be difficult for word learners. However, the difficulty may depend on the meanings: prior work has demonstrated that word-learning is easier for both adults and children when words' multiple meanings are related (polysemous, like "cap”) than unrelated (homophonous, like "bat”). However, it remains an open question how learners infer polysemous meanings if learners encounter these words in more referentially ambiguous contexts. In two studies, we examine children’s and adults’ learning of polysemes under uncertainty, using both artificial stimuli from prior work (Study 1) and attested non-English polysemes (Study 2). Results suggest that while adults can use similarities between referents to infer polysemous meanings across multiple exposures, children generally struggle to do so. This indicates that polyseme learning improves with age and suggests current computational models of cross-situational word-learning may capture children’s word learning strategies better than those of adults.