How spatial information connects visual perception and natural language generation in dynamic environments: Towards a computational model

Wolfgang Maaß

In: Andrew U. Frank, Werner Kuhn (Hrsg.). International Conference on Spatial Information Theory. Conference on spatial information theory (COSIT-95) September 21-August 23 Semmering Austria Seiten 223-240 Springer 1995.


Suppose that you are required to describe a route step-by-step to somebody who does not know the environment. A major question in this context is what kind of spatial information must be integrated in a route description. This task generally refers to two cognitive abilities: Visual perception and natural language. In this domain, a computational model for the generation of incremental route descriptions is presented. Central to this model is a distinction into a visual, a linguistic, and a conceptual-spatial level. Basing on these different levels a software agent, called MOSES, is introduced who moves through a simulated 3D environment from a starting-point to a destination. He selects visuo-spatial information and generates appropriate route descriptions. It is shown how MOSES adopts his linguistic behavior to spatial and temporal constraints. The generation process is based on a corpus of incremental route descriptions which were collected by field experiments. The agent and the 3D environment are entirely implemented.

Weitere Links

how-spatial-information-connects-visual-perception-and-natural-language-generation-in-dynamic-environments-towards-a-computational-model.pdf (pdf, 233 KB )

Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence