
DYNAMIC PAGE -- HIGHEST POSSIBLE CLASSIFICATION IS
TOP SECRET // SI / TK // REL TO USA AUS CAN GBR NZL
(U) Machines Translating Arabic
FROM: SIGINT Communications
Run Date: 01/21/2004
FROM: SIGINT Communications
(U) NSAers are participating in a study to answer the question: "How good must Arabic machine
translation tools be to be operationally useful?"
(U) Since 2001, the U.S. Government's need for Arabic translation skills has surged
dramatically. With military operations in Afghanistan and Iraq ongoing, and Arabic linguists in
high demand, the government began several projects to develop an effective Arabic machine
translation (MT) tool. NAIC, Ft. Huachuca, CIA and DARPA all put funds into separate efforts.
Government officials soon realized that it would be smarter to concentrate the funding on the
one or two most promising projects, instead of scattering the money among several. The
question remained, however: which are most promising?
(U) To get to the bottom of this question, the Center for Advanced Study of Language (CASL),
formed last year at the University of Maryland, was tasked with conducting a task-based
evaluation of some of the Arabic MTs that were funded in 2003. (This evaluation was proposed
by John Kovarik, NSA's Senior Language Technology Authority.) Such a study is now underway.
With the help of NSA analysts, CASL will study how well three representative MTs -- each
representative of a different approach -- work for analysts who are not familiar with Arabic.
(U) The study will work like this: unclassified documents will be sent through the three systems.
The output will then be randomly presented to NSA analysts in such a way that they will not
know which system produced which output. The analysts will then read the documents and try
to answer various questions, based on what they have read. Software will grade the answers
and track how long the analysts take to answer the questions. These "grades" will then be
summed to measure how well the three MTs work.
(U) NSA analysts are currently being sought to volunteer to participate in the study.
Participants will be trained and take the "tests" at UMd on January 26. If you are interested in
volunteering, please contact John Kovarik (
nsa;
) or Lynn Carlson (Senior
Technical Director of CASL's Government Oversight Board), externally at CASL:
@casl.umd.edu;
"(U//FOUO) SIDtoday articles may not be republished or reposted outside NSANet
without the consent of S0121 (DL sid_comms)."
DYNAMIC PAGE -- HIGHEST POSSIBLE CLASSIFICATION IS
TOP SECRET // SI / TK // REL TO USA AUS CAN GBR NZL
DERIVED FROM: NSA/CSSM 1-52, DATED 08 JAN 2007 DECLASSIFY ON: 20320108