Automated Text Highlighting of Navy Equipment Failure Messages
Abstract
This report describes Text Reduction System (TERSE) that is a knowledge-based system for highlighting important information in the narrative portion of Navy equipment failure messages-Casualty Reports (CASREPs). The system contains two knowledge bases for message evaluation, one that is equipment-specific and the other equipment-general. The equipment-specific knowledge base contains a structural model of a piece of equipment discussed in one class of CASREPs (a shipboard air compressor), encoded as a network of slot/ filler units. Since message writers use a wide variety of descriptive naming conventions in referring to pieces of equipment, it is not possible to provide a complete list of synonyms for each part. Instead, the system must use the equipment model to actively de-reference each complex nominal, by finding an equipment unit whose attributes match a structural host-modifier analysis of the noun phrase. When an equipment name is underspecified, a disambiguation algorithm uses the equipment model to select the most likely referent from the ambiguity set of matching units. The system also contains general causal inferencing heuristics that use the equipment model network to infer causal relationships that are believed to be implicit in the message. The equipment- general portion of the system performs semantic normalization, infers and tags key categories of information, and finally ranks the message clauses by applying user evaluation criteria represented as numeric scores assigned to various patterns of information types. The system implemented in the KEE expert system shell and runs on a Symbolics machine and Sun workstation.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 02, 1988
- Accession Number
- ADA202961
Entities
People
- E. Marsh
- K. Wauchope
- M. K. Dibenigno
Organizations
- United States Naval Research Laboratory