The Army will upgrade Pathfinder, a widely used but two-decade-old intelligence analysis system.
The Army is going to give Pathfinder a facelift, officials said. The widely used but two-decade-old intelligence analysis system processes, integrates and analyzes data from government, commercial and public domain databases.
Tim Estes, president of Nashville-based Digital Reasoning Systems, said the company is developing a computational linguistics software product, Interceptor, which can process language patterns to produce better quality analysis of structured and unstructured data.
Earlier this year, the National Ground Intelligence Center (NGIC), the Army’s intelligence analysis organization based in Charlottesville, Va., awarded Digital Reasoning Systems a sole-source contract to upgrade systems like Pathfinder. Additionally, NGIC officials negotiated a license to upgrade Pathfinder with Interceptor.
NGIC has worked with the company in some capacity since spring 2001. It is planning to exhibit some features of Interceptor this fall for a few agencies and organizations. The company may need a year to install Interceptor at nearly 50 organizations with 15,000 Pathfinder users.
The military created Pathfinder in the 1980s to provide intelligence analysts with automated tools to compare, link, access and visualize large amounts of data. Before the system, analysts had to manually perform such tasks.
Although Pathfinder is five to 10 times more prolific than any other government system, Estes said, the upgrade will be like putting Windows on DOS.
Interceptor, he said, not only gives analysts a search function, which it does very well, but also automatically maps unstructured data or human language into relationships. It learns associations and correlations between different linguistic items.
For example, Estes said, the analysis software can figure out that boats are associated with water and boats can have oars or motors, depending upon the data. It “builds up a set of relationships that starts to look almost ontological,” he said.
He said Interceptor works like the human brain, which learns new information and references data already encountered. Therefore, the system only produces data that is new for analysts.
For example, “the amount of new information on the Internet is small,” Estes said. “How many blogs…have postings that just pick up someone else’s and say the same thing over again?
“Now there’s a lot of data that is like that and even intelligence data can be like that,” he added. “So the question is, ‘Do you want to reinforce what you have already learned, or do you want to say this is something new?’”
Estes said other intelligence analysis applications are little more than search engines with added functions, such as categorization or document grouping. But the other products don’t create relationships.
Typically, those solutions try to model various permutations of language by using dictionaries, encyclopedias and other reference materials, he said. But when words’ meanings change, their systems break, he said. They’re “not really learning based on how language is used, they learn language based on looking it up,” he said. “So they don’t learn at all.”
For example, Estes said, an enemy might use “chocolate cake” as code to mean some kind of precursor to an explosion. In a typical system, chocolate cake would have no importance. But with Digital Reasoning Systems’ software, chocolate cake might be associated with certain people considered enemies and with an event like a bombing.
Interceptor does not start with a model in advance. As data enters the system, it creates patterns of association. Estes said that means the system essentially evolves because there is an order in the meaning of language “that is not just a dictionary in the sky.”