Much work has documented children's comprehension difficulties with verbs of cognitive state--e.g. "think", "believe", "realize", "want", "say", etc. Adult-like representations of desire and communication verbs like "want" and "say" and belief state verbs like "think" do not appear to be present until 3 and 4 years of age, respectively (de Villiers 2005). One factor in this late-learning may be the nature of the evidence that children have at their disposal for learning these verbs. Due to the essential unobservability of cognitive states, linguistic context must play a major role in children's learning the meaning of cognitive state verbs (Gleitman 1990). Specifically, the types of sentences a verb occurs in can help a learner shrink the space of possible meanings for that verb. Here I explore the question how much information the frame distributions for different subcategories of cognitive state verbs provide, with results from two studies.