University of Massachusetts: MUC-4 test results and analysis

The UMass/MUC-4 system is based on a form of sentence analysis known as selective concept extraction. This approach to language processing is distinguished by a minimal reliance on syntactic sentence analysis, along with a minimal dictionary customized to operate in a limited domain. Last year, the UMass/MUC-3 system demonstrated the viability of selective concept extraction, but serious questions were raised about the portability and scalability of the technology, particularly with respect to the creation of domain-dependent and task-dependent dictionaries. We estimated that 9 person/months went into the creation of the dictionary used by UMass/MUC-3, and we were unable to say how much domain-dependent lexicon was still missing. We were nevertheless sure that our dictionary coverage was incomplete.