Children can learn words by tracking cross-situational co-occurrences between words and referents. However, children also use information about speakers’ mental states to support word learning, a process not captured by purely associative learning mechanisms. In two studies, we asked how statistical and epistemic cues act together during word learning, and whether sensitivity to speaker knowledge helps children restrict cross-situational associations between words and objects. 
In Study 1, children aged 3.5–6 and adult controls completed a cross-situational word learning task that manipulated speaker knowledge through statements of ignorance. We found that both adults and children reasoned about speaker ignorance to restrict possible meanings of novel words. In Study 2, we used the same paradigm, and manipulated speaker knowledge through visual access. Adults reasoned about the speaker’s visual perspective to learn words, but we did not find conclusive evidence that children did so. These results suggest that young children can use information about speaker knowledge to constrain cross-situational word learning, but only when the epistemic cues are clear and explicit.