ARLINGTON, Va., 27 May 2012. Intelligence experts at the U.S. Defense Advanced Research Projects Agency (DARPA) in Arlington, Va., are asking industry to develop computer algorithms and a computer system to help intelligence analysts pull useful, actionable information out of large amounts of text.
DARPA issued a broad-agency announcement (DARPA-BAA-12-47) last week for the Deep Exploration and Filtering of Text (DEFT) program, which seeks to develop text-analysis computers and algorithms to interpret ambiguities in natural language and developing actionable intelligence.
The DEFT program will develop the ability to see through language to meaning in text, identify key information in text documents, identify sources that contain important new, and to automate the initial stages of report writing.
The goal of the program is to create an end-to-end system that will assist the warfighter and intelligence analyst at every stage -- from searching through data to identifying threats and then formulating analyses.
Military analysts collect and process copious amounts of data from a wide range of sources, DARPA officials explain. To help them, DARPA wants the ability extract relevant information automatically that is only referenced indirectly and not be readily apparent to humans.
Text properties that challenge automated language processing systems include people and groups; events; geo-spatio-temporal information; causal explanations; facilities and equipment; modality and beliefs; anomaly, novelty, emerging trends; inter-relationships and entailments; disfluencies and disjointedness; dynamic, perishable, changing situations; ambiguity and vagueness; uncertainty, inconsistency, contradiction; and unstated, implied information.
The DEFT program aims to enable analysts to discover implicitly expressed, actionable information from ambiguous and vague narrative text. DARPA is organizing the program over 4.5 years in three 18 month phases. The program has three technical areas of interest: algorithm development; systems integration and deployment; and data creation and annotation.
Program goals include developing algorithms that handle real-world input, and integrating these algorithms into a DEFT System for deep natural language processing; and deploying this capability to users.
Smart filtering is to reduce the amount of material that analysts must read by collapsing or clustering overlapping or redundant information, and will rely on algorithms for language redundancy and semantic equivalence.
Ultimately the DEFT System should create alerts of possible new developments, as well as a pre-report summary of points of interest that an analyst may want to put into a report or into a model.
Companies interested should respond to DARPA no later than 20 Aug. 2012. For questions or concerns, contact Bonnie Dorr, the DEFT program manager, by e-mail at DEFT@darpa.mil.
More information is online at https://www.fbo.gov/spg/ODA/DARPA/CMO/DARPA-BAA-12-47/listing.html.