By the time children begin to rapidly acquire new word meanings they are already able to determine the grammatical category of novel words based on syntactic and morphological cues. Here we test whether children can leverage this knowledge when inferring the meaning of a novel word. Through a novel word learning experiment we determine that children can use this information, drawing different conclusions for the most likely meanings of novel words in distinct grammatical categories. We use a Bayesian model to formalize the higher level knowledge that children might have about noun and adjective meanings. Simulations show that children's behavior reflects the type of shift we would predict on the basis of noun and adjective meanings in the English lexicon.