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

Abstract

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 staxting-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 axe entirely implemented.