Impact of Machine-Translated Text on Entity and Relationship Extraction

Report No. ARL-TN-0649
Authors: Mark R Mittrick; John T Richardson
Date/Pages: December 2014; 28 pages
Abstract: We performed an experiment to study the effects of machine (performed by software) versus manual (performed by a human) translation on the performance of a Small Business Innovation Research text analytics tool. The text analytics in the experiment is Contour, developed by Decisive Analytics Corporation, which automatically builds high-fidelity social networks from text data sets too large to be scrutinized in detail through manual effort. Specifically, we analyzed the ability to extract text entities with the roles of person, location, or organization. The data consists of the translations of many news stories collected from Arabic language websites. There are 5 translations for each story to examine (4 human and 1 machine). The performance of the machine translation Contour results is analyzed against the Contour results of the manual translation.
Distribution: Approved for public release
  Download Report ( 0.875 MBytes )
If you are visually impaired or need a physical copy of this report, please visit and contact DTIC.

Last Update / Reviewed: December 1, 2014