Language Systems, Inc.: description of the DBG system as used for MUC-3

LSI's Data Base Generation (DBG) approach to natural language understanding is characterized by three main features: First, the DBG system is comprehensive. It performs full-scale lexical, syntactic, semantic, and discourse analyses of the entire message text being processed and produces a complete knowledge representation of the text. Second, the DBG system is modular and flexible. The modular and transparent system architecture ensures easy extension, maintenance, and upgrading of the system. And, third, the DBG system is generic but at the same time domain-sensitive. It applies domain modeling to text interpretation, which enables the extension of the system to any number of new domains. In addition, it provides a powerful capability for handling unknown data in familiar domains. DBG's development has been based on analysis of large volumes of message traffic (thousands of Air Force and Army messages) in five domains, as described below. The system can currently process a large number of messages in each of these domains and has been formally tested on previously unseen messages in three of these, with competitive tests against humans performing the same task in two domains. The functional flow of the DBG system is shown in Figure 1 (actually Figure 1 of our site report [Language Systems Inc: MUC-3 Test Results and Analysis] in this proceedings).