Leveraging word knowledge from everyday language

Within the first few years of life, children go from nonverbal infants to fluent users of hundreds of words. This feat unfolds as children are immersed in a sea of words in conversations, stories, and other language contexts. In principle, word contexts might support word learning because words with similar meanings tend to occur in similar contexts. Thus, a child who knows some fruit words and has heard them in the context of "juicy" might learn that a "juicy mango" is also a fruit even without ever seeing a mango. Indeed, cutting-edge large language models leverage just these context cues to form rich representations of words. Yet we do not know whether these cues are an important driver of real-world word learning. We quantified context cues in children's everyday English language input and found that they did reliably predict real-world word learning, above and beyond other established predictors. These findings illuminate how everyday language scaffolds word learning.