Visualization of Narrative Events
Document Actions
- Speaker:
- Pierre Nugues, Department of Computer Science
- Abstract:
- In this presentation, we will introduce and discuss syntactic and semantic components to visualize events described in a narrative. Ultimately, these components would serve as building blocks to develop automatic converters of textual descriptions into animated 3D scenes. These components are a generalization of those used in an existing prototype, Carsim, which is applied to road accidents written in Swedish.
Carsim takes a written report relating an accident as input, constructs a 3D scene from it, recreates the vehicles, and animates them on a computer screen. The prototype enabled us to design a flexible and extensible model of the knowledge contained in the texts. We created an architecture that combines language processing and visualization techniques for which we obtained satisfactory results for road accidents. However, Carsim is domain dependent.
In our presentation, we will concentrate on the language processing aspects of text-to-scene conversion. We will describe the syntactic and semantic modules we have developed for Carsim and their extensions to make them generic. Although not always universal, most language processing modules are fairly reusable and applicable to other domains. Potential application areas are story telling, instruction manuals, and other kinds of accidents.
Last modified 2006-09-20 14:03
