02/01/1992 09:00 AM
Computer Science
If a user were to imagine the most convenient method of manipulating objects, it would probably be language-based. However, natural language understanding is notoriously difficult for computers, and the difficulty is exacerbated in the case of computer graphics by the need for geometric output, not just ``conceptual\'\' understanding or high-level inferencing. General text understanding techniques have not been successfully generally applied to scene generation; more typically, a few task-specific commands, such as ``walk\'\', are implemented in detail and independently. We have chosen a more generalizable, bottom-up approach of developing a small natural class of commands useful for object manipulation. We believe this is a first step toward a much more natural method for object manipulation using computer graphics.
