. German Research Center for Artificial Intelligence GmbH
Within the project VITRA (VIsual TRAnslator) we are concerned with the design and construction of integrated knowledge-based systems capable of translating visual information into natural language descriptions. In this contribution the focus will be on high-level scene analysis, i.e., the step from a geometrical representation, as it might be provided by a vision component, into conceptual descriptions of object motions. In a combined vision and natural language system aiming at simultaneous natural language descriptions of dynamic imagery the recognition of such motion events has to be done incrementally in order to be able to talk about events even while they are progressing. We present course diagrams as a means for such an incremental motion analysis and show how they can be constructed automatically from interval-based event representations using temporal constraint propagation techniques.