The process of verb learning in children involves learning the verb's syntactic structure and also learning how to relate this structure to corresponding semantic knowledge. For example for the English transitive frame describing an action involving two participants, the child has to learn how the two participants are distinctively referred to by the subject and the object of the transitive verb. This knowledge of the mapping between syntactic and semantic roles is a critical part of any theory of language acquisition; yet little is known about its nature and its origin. The traditional view is that children possess a built-in abstract notion of the semantic roles as a prerequisite to language. This thesis argues for an alternative view, one in which knowledge of roles emerges hand-in-hand with knowledge of verbs on a case-by-case basis.

The emergence of the semantic roles known as Agent and Patient was studied in a series of experiments. Experiment 1 used novel objects and actions to examine the generalizations made by 3-to-5-year-old children between Agents and Patients, with and without the benefit of language. The findings revealed that linguistic cues are important for directing children's attention to the properties shared by Agent-Patient roles for both age groups, but particularly among younger children. Experiment 2 explored two key components used to identify Agent-Patient roles: change-of-state and contact. The results showed differential effects of the absence of each component on children's ability to identify Agent-Patient roles in novel action scenes. Together, the findings are consistent with the hypothesis that children do not start with abstract knowledge of Agent/Patient roles.

A computational model was constructed to simulate children's behavior in the experiments. Utilizing a case-based approach, the model starts with no hard-wired knowledge of roles and learns how to find the Agents and the Patients of the novel actions through exposure to many cases of similar actions sharing the same verb argument structure and relational semantic features. The model provides converging evidence that children's knowledge of roles develops as a result of exposure to language and prior experience with the world, and sheds new light on the mechanisms by which children acquire the syntactic knowledge of verb argument structure.