A Marker Propagation Text Understanding and Inference System

Introduction This paper describes a natural language understanding system that is under development at the Southern Methodist University. The project, known as the SNAP project, started at the University of Southern California in late 1980’s. Our group participated in the Message Understanding Conferences (MUC-4 and -5) where the task was to extract information from thousands of real messages and fill in some predetermined templates. In this paper we describe the understanding and inference components of our system. A distinct feature of our system and an important innovation is the application of parallel processing to natural language. Another significant aspect is that we take a complete system approach that integrates many levels of language processing and aim at understanding unrestricted text. To facilitate the processing of unrestricted text, we use a knowledge base constructed on top of WordNet 1.5, the lexical database for English language developed at Princeton[Mil95].