Generating Semantic Information for Virtual Environments

In order to build virtual humans with pleasing interactions in virtual scenarios, a consistent representation of all relevant relations and information is required. For virtual scenarios, this equates to having a meaningful (semantic) understanding of the objects and virtual agent performable actions in the scenario. The process of creating semantics for rich virtual environments is a tedious and time consuming endeavor, such that only the most pertinent pieces of information are added to the most used objects in the scene. This has the effect of decreasing realism as only a few objects can be acted upon. Furthermore, authoring relations by hand does not guarantee consistent semantics, which can also erode realism. We discuss the need and progress in automating semantic generation for objects and actions in virtual scenarios. We provide special consideration for semantics connecting objects to actions, known as operational information. We also examine the need for automated methods to generate domain specific knowledge such that a virtual environment is not overburdened by semantic information but instead finds a balance of relations.