Technology aids hunt for terrorists

  • Dan Verton (Computerworld)
  • 09 September, 2002 08:15

Analysts and field operatives from the U.S. Central Intelligence Agency (CIA) and the National Security Agency (NSA) have stepped up offensive operations against terrorists around the world using sophisticated text and audio search and analysis technologies.

"We've increased the number of teams around the world collecting information and disrupting [terrorist] activities," acknowledged Philip Lago, executive secretary of the CIA. "The tempo of that activity has increased dramatically."

Those operations have also increased the volume of raw technical intelligence -- phone, radio and video recordings as well as text -- flowing into the headquarters of the CIA and NSA in Langley, Va., and Fort Meade, Md., respectively. The result: a dramatic upswing in demand for technology to help ensure that analysts don't miss critical communications or code words that could be used to launch an attack.

One of those technologies is the Name Reference Library from Language Analysis Systems Inc. (LAS) in Herndon, Va. The software analyzes name origins, tells the user whether or not multiple middle and last names are in the right order (Egyptian- and Saudi-born citizens often use multiple generational names), and provides a list of the top 10 spelling variants as well as gender associations.

LAS is working on a product that will enable processing of native scripts, said Jack Hermansen, the firm's president. "If you can capture Mohamed in Arabic, for example, it's only spelled one way. The problem is in the transcription to other languages," he said. LAS plans to complete development work on the new version in the next six months.

Meanwhile, NSA analysts, who are responsible for intercepting and analyzing hundreds of terabytes of archived and real-time voice, data and video communications, are getting help from Fast Talk Communications Inc.

Fast Talk President Armistad Whitney said the company's software can break down speech to its smallest components, called phonemes. The phonemes can then be indexed and searched for keywords. The software can retrieve any word, name or phrase from voice data, regardless of speaker or dialect, with up to 98 percent accuracy and up to 72,000 times faster than in real time, company officials said. Analysts can therefore search through 20 hours of audio in less than 1 second.

The company signed two contracts with the intelligence community within the past 90 days. Currently in the deployment phase, the software is being loaded on laptop computers for use in the field and on multiprocessor enterprise systems at agency headquarters, Whitney said.