<?xml version="1.0" standalone="yes"?>
<Paper uid="W01-1301">
  <Title>Generating a 3D Simulation of a Car Accident from a Written Description in Natural Language: the CarSim System</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> This paper describes a prototype system to visualize and animate a 3D scene from a written description. It considers the narrow class of texts describing car accident reports. Such a system could be applied within insurance companies to generate an animated scene from reports written by drivers. The research is related to the TACIT project (Pied et al., 1996) at the GREYC laboratory of the University of Caen and ISMRA.</Paragraph>
    <Paragraph position="1"> There are few projects that consider automatic scene generation from a written text, although many projects exist that incorporate natural language interaction in virtual worlds, like Ulysse (Bersot et al., 1998; GodPereaux et al., 1999) or AnimNL (Badler et al., 1993). Visualizing a written car accident report requires a difierent approach.</Paragraph>
    <Paragraph position="2"> It is closer to projects focusing on text-to-scene conversion, like WordsEye (Coyne and Sproat, 2001). However, unlike the latter, our objective is to build an animation rather than a static picture and behavior of dynamic objects must then be taken into account. There also exist systems that carry out the reverse processing, from video data to text description, as ANTLIMA (Blocher and Schirra, 1995).</Paragraph>
    <Paragraph position="3"> We present here an overview of the CarSim system that includes a formalism to describe and represent car accidents, a linguistic module that summarizes car accident reports according to this formalism, and a visualizing module that converts formal descriptions to 3D animations. In our case, the linguistic module has to deal with texts where syntax and semantics involve time and space description and simultaneous actions of two or more actors (i.e. the cars).</Paragraph>
    <Paragraph position="4"> The remainder of our paper is organized as follows. Section 2 presents the formalism for describing an accident. Section 3 describes the template fllling methods that lead to the conversion of a text into its formal representation. Section 4 covers planning techniques and accident modelling algorithms that we use. Finally, Section 5 presents and discusses the evaluation of the system on the test corpus (MAIF corpus).</Paragraph>
    <Paragraph position="5">  mal Description) as a means of communication.</Paragraph>
  </Section>
class="xml-element"></Paper>