This thesis is part of a multidisciplinary project, the DRAMA project, which attempts to generate 3D virtual scenes from the descriptions which are obtained from theatrical text. This project aims to simplify, as soon as possible, the tasks of the end-users by providing simple, fast, and effective tools. Thus, the technique used in this study is focused on the declarative modeling of virtual environments that is based on three phases (description, generation and management of knowledge). The description phase allows the designer to describe the environment from a set of properties, interpreted as a set of constraints for a generation system which produces one or several virtual environments solutions. This project, new tagging methods have been proposed to detect essential for the creation of scene, including information on the placement of objects. In addition, users can also run queries in the text from these tags. Placement properties are translated into spatial constraints with the data originally stored in a knowledge base that uses XML. A technique adopting the method of metaheuristics is then used for solving constraints. The object physical properties (collision, gravity, friction) were also managed from a physics engine. At the end, the finals scenes solutions were be proposed to the user, using a 3D rendering engine.
