Communication among agents in a natural language places certain requirements on the conceptual model, and language itself places constraints on what can be communicated fluently, without circumlocution. In this paper, I explore the relationship between conceptual modeling and natural language generation, focusing on the problem of representing and describing events. A text generator producing the linguistic representation of actions and objects, either as part of the communication among agents or a description of their actions, has its own set of requirements for what information needs to be represented in order to map from the model to linguistic form. I describe six kinds of information needed for the expression of events: linear time, event type, temporal modifiers, event structure, and agency and discuss whether that information best reside in the underlying model or in the linguistic component. The discussion is grounded in a particular system, SAGE, which includes an event based sinmlator, knowledge base, and text generator. However, the issues are the same whether the underlying model drives a simulator, is the output of vision system or some other program.