People don’t always say what they mean, especially when they know they are being watched.
Superfluous language and cultural nuance often prevent defense analysts from making the right connections when poring through documents containing intelligence gathered during military missions. Many times, information is deliberately obscured and important matters are addressed in roundabout ways.
The Pentagon wants to get to the bottom of what it calls “deep natural language” by creating an automated system that can process text at its most basic level to reveal meaning that otherwise may not be apparent.
Analysts have mountains of information to go through and never-ending deadline pressures. In this stressful environment, they can miss crucial links, said Bonnie Dorr, who manages the Deep Exploration and Filtering Text (DEFT) program for the Defense Advanced Research Projects Agency.
“Especially when meaning is being deliberately concealed or otherwise obfuscated,” she said. “We want the ability to mitigate ambiguity in text by stripping away filters that can cloud meaning and by rejecting false information.”
To be successful, the technology must look beyond what is explicitly expressed in text to infer what is actually meant. It would allow analysts to take a more strategic approach to exploring information. Currently, they try to process insurmountable quantities of data in a limited, linear fashion, officials said.
DARPA hosted a “proposers’ day” workshop for the program in May, after which officials planned to issue a broad agency announcement to solicit proposals for industry participation.