Architecture of Spatial Ontology for Text-to-Scene Conversion System

Spatial concepts conversion in text-to-scene conversion systems (TTS) is one of the basic tasks. Spatial concepts conversion is to transfer the object locations specified with natural language into the objects display in the virtual scene. In the present, there are some methods including image schemes, expert knowledge base and graphic annotation to resolve the spatial concepts conversion in TTS. The problems of these methods are human labor consumed and skilled art training needed. In this paper, an architecture of spatial ontology for domain of text-to-scene conversion is provided, which makes up the gap between natural language system and graphic system from spatial concept point of view and improves the acceptability of the output scene. The architecture of domain spatial ontology is designed on the domain spatial ontology modol in the framework of SUMO top level. The architecture has two levels, one is the top derived from SUMO top level and has 8 nodes, and the other is the low defined on the features of TTS domain and composed of subclasses of nodes of SelfConnectedObject, Relation and Attribute from the top level. The spatial ontology is tested in the experiment of objects displayed on Chinese text in the text-to-scene system and a good result is gotton.