Computational aspects of discourse in the context of MUC-3
MUC3 '91 Proceedings of the 3rd conference on Message understanding
University of Massachusetts: description of the CIRCUS system as used for MUC-4
MUC4 '92 Proceedings of the 4th conference on Message understanding
Classifying texts using relevancy signatures
HLT '91 Proceedings of the workshop on Speech and Natural Language
Classifying texts using relevancy signatures
AAAI'92 Proceedings of the tenth national conference on Artificial intelligence
| Hi-index | 0.00 |
This report contains a technical description of a sophisticated natural language processing system designed to operate on unconstrained texts describing acts of terrorism. This partic- ular system was used in the final evaluation of the Third Message Understanding Conference (MUC-3), where it posted the highest recall rates and highest combined scores for recall and precision of all the systems under evaluation. Details are provided with respect to semantic and syntactic sentence analysis as well as the discourse-level analysis of complete texts. Of particular interest is our ability to robustly analyze texts using only a minimal dictionary (6,000 words), and without generating syntactic parse trees for individual sentences. A case- based reasoning (CBR) approach to discourse analysis is also noteworthy. In addition, this system description indicates how much domain-specific effort was needed to bring a generic sentence analyzer up to speed for a particular information extraction application.