Understanding scene descriptions by integrating different sources of knowledge

The aim of this work is to describe a system, called NALIG (Natural-Language-driven Image Generator), able to understand natural language descriptions of object spatial configurations and draw on a graphic screen one of the infinitely many scenes consistent with the input. NALIG can be used interactively and at any step a multi-layered contextual analysis of the input is performed to detect and eliminate possible incosistencies due to previous default choices. The system is described in its functionalities, internal structure and interactions among its sub-parts.