An Examination of Training Effects on Entity Extraction Systems in Manually Translated Iraqi Military Documents

Report No. ARL-TR-6159
Authors: Michael H. Lee and Robert P. Winkler
Date/Pages: September 2012; 20 pages
Abstract: An Entity Extractor may perform poorly if it is used on input text that differs from its trained domain. This report describes the results of running City University New York's Information Extractor on English translated military domain Iraqi documents and how the results were improved with various retraining methods. Improvements in precision, recall, and F-Score are visually presented.
Distribution: Approved for public release
  Download Report ( 0.629 MBytes )
If you are visually impaired or need a physical copy of this report, please visit and contact DTIC.

Last Update / Reviewed: September 1, 2012