Creating VR Simulations Using Natural Language Human instructions come in a variety of forms (e.g. checklists, flowcharts, procedural instructions, etc.). Creating computer graphics simulations of humans performing instructed tasks, often involves programming and at best a graphical user interface (GUI) that does not correlate well with the form of instructions given to the graphical figure's human counterpart. This makes creating simulations a time consuming special skill, particularly when it is difficult to reuse animations or behaviors from one simulation to another. The Center for Human Modeling and Simulation is constructing a software simulation environment that addresses these issues through a parameterized action representation (PAR), a persistant database of actions, and a GUI that combines standard GUI elements such as menus, buttons, and dialog boxes with the ability to enter natural language instructions. The PAR was designed as a natural language, computer graphics simulation intermediary. It contains parameters that reflect the semantics of natural language terms that can be used in artificial intelligence level planning. It also contains parameters about the movements of the actions, such as dynamics and constraints, which are used to depict the actions in a graphical simulation. As the particulars of an actions are often dependent on the objects and agents participanting in the action, PAR also contains representations of objects and agents, ranging from their utilities or capabilities to grasp sites on the objects. PAR actions are linked to motion generators that are capable of displaying the actions on a graphical figure in a 3D simulation. PAR object representations can also be linked to the data files containing the geometry of the objects. PAR actions and objects are stored in a persistant database called the Actionary. In effect, these actions and objects are primitives that can be combined in order to create simulations for trainning, human factors evaluations, instruction validation, and interpersonal crew capatability analysis. We are currently creating PARs within an XML framework, so that they can more easily be shared with other research groups. The current focus of our work is also designing an interface to the Actionary that will best faciliate a naive user's ability to create a new simulation. Natural language instructions alone are often too ambigious to be used to create a simulation, but when combined with traditional GUI primitives and added information and suggestions generated by information stored in the Actionary's actions and objects, we believe no specialized computer expertise will be needed.