Carsim: A System to Convert Written Accident Reports into Animated 3D Scenes

This paper describes a system to create animated 3D scenes of car accidents from reports written in Swedish. The system has been developed using news reports of varying size and complexity. The text-to-scene conversion process consists of two stages. An information extraction module creates a tabular description of the accident and a visual simulator generates and animates the scene. We first describe the overall structure of the text-to-scene conversion and the template structure. We then explain the information extraction and visualization modules. We show snapshots of the car animation output and we conclude with the results we obtained. 1 Text-to-Scene Conversion Images and graphics have an indisputable capacity to represent and to communicate knowledge, see [1], [2], and [3]. As it has been frequently noted, it is often easier to explain physical phenomena, mathematic theorems, or structures of any kind using a drawing than words. Images can help understand ideas or situations and realize their complexity. They are an effective means to describe and explain things. Text-to-scene conversion consists in creating a 2D or 3D geometric description from a text and in displaying it. The scene can be static or animated. To be converted, the text must be appropriate in some sense, that is, contains explicit descriptions of objects and events for which we can form mental images. The conversion of natural language texts into graphics has been investigated in a few projects. NALIG [4] [5] is an early example of them that was aimed at recreating static 2D scenes. One of the major goals of the project was to study relationships between space and prepositions. NALIG considered simple phrases in Italian of the type subject, preposition, object that in spite of their simplicity can have ambiguous interpretations. From what is described in the papers, NALIG has not been extended to process sentences and even less to texts. WordsEye [6] is a recent and impressive system that recreates 3D animated scenes from short descriptions. The number of 3D objects WordsEye uses – 12,000 – gives an idea of its ambition. WordsEye integrates resources such as the Collins’ dependency parser and the WordNet lexical database. The interpretation of a narrative is based on an extension of case grammars (semantic frames) and a good deal of inferences about the environment. WordsEye does not address real world stories. The narratives cited as examples resemble imaginary fairy tales. In addition, all the cited texts appear to have been invented by the authors. CogViSys is a last example that started with the idea of generating texts from a sequence of video images. The authors found that it could also be useful to reverse the process and generate synthetic video sequences from texts. The logic engine behind the text-to-scene converter [7] is based on the Discourse Representation Theory. The system is limited to the visualization of single vehicle maneuvers at an intersection as the one described in this two-sentence narrative: A car came from Kriegstrasse. It turned left at the intersection. The authors give no further details on the text corpus and no precise description of the results.