Abstract
The model to implement the process of the decoding of a spatial description to its spatial image -- topology of the adjacent relations - has been created. The model enables the implementation of the system that updates and maintains the spatial model(s) constructed from the discourse. The model is trying to solve the four problems concerning how to resolve the reference of an anaphoric expression, how to handle transition of the utterer's point of view, how to detect and eliminate inconsistent information, and how to understand implicit and/or incomplete descriptions. The solution relies on the four assumptions derived from the literature survey. (1) Interaction between the system and the user helps the system to understand the meaning and reference of an ambiguous or vague description. (2) The transition of a point of view is calculated by tracing the point and perspective of the effective point of observation. (3) Consistency of the spatial relations is maintained by the combination of a spatial reasoning algorithm and the dual graph representation formalism and its rewriting rules. (4) Incomplete and/or implicit description is complemented based on knowledge about spatial relations and architecture.
