Children rapidly learn thousands of words and a productive language system that allows them to say and understand sentences they have never heard before. One challenge that children face in acquiring this complex web of linguistic knowledge is to link words to their referents. Adult listeners rely both on their knowledge about linguistic structure as well as their knowledge about events to determine reference. In this talk, I will show that children do the same. Specifically, I demonstrate that by their 3^rd birthday children generate causally structured mental models of described events and leverage these models to support referential inferences during online processing. I additionally demonstrate that this ability depends on and reflects language skill. This work suggests that children understand language not only by learning about statistical patterns in language use, but also by generating rich conceptual representations of what has been said; these representations can then be accessed rapidly as sentences unfold to support continuous, incremental comprehension.