Can DARPA Teach Machines to Read? (updated)

The Defense Advanced Research Projects Agency (DARPA) has launched the Machine Reading program to develop a revolutionary, automated reading system that bridges the gap between naturally occurring text and the artificial intelligence (AI) reasoning systems that need such knowledge.
AI systems continue to grow in use by the US military as there is a consistent emphasis on using high technology as a strategic advantage and reducing reliance on humans.
DID has more on this military application of AI and a recent IBM contract…
Situation awareness, diagnostics, prognostics, planning, logistics – all are areas in which AI systems are used in military applications. A great deal of the militarily relevant knowledge that these systems need is presently expressed as natural-language text, unusable for AI systems. This knowledge may range from information about local political and militant groups to infrastructure and food supplies. The DARPA program intends to enable AI systems to use the vast amount of information available in text formats.
Although this intelligent learning system would initially be used for military purposes, it could also enable a variety of civilian applications. For example, as more and more of the world’s libraries are converted to digital text, the system could provide unprecedented analysis using AI systems of this vast storehouse of knowledge.
DARPA is working with the U.S. Air Force Research Lab (AFRL) in Rome, NY on this effort. AFRL recently awarded a $23.7 million contract (FA8750-09-C-0172) to IBM Corp. to develop a prototype machine reading system that builds domain knowledge automatically from input text.
In addition, AFRL awarded a $29.7 million contract (FA8750-09-C-0179) to BBN Technologies to develop a universal text engine that captures knowledge from text and transforms it into the formal representations used by AI systems.
A central goal of the research effort is to develop techniques that can generalize across the linguistic structure and content of documents to extract relations and axioms directly from text, rather than relying on a person to encode such information. A related goal is to develop techniques capable of performing automatic extraction of text on the Web. Over the course of the 5-year program, BBN’s system will be tested against increasingly complex targets, including its ability to learn axioms from text and to read and digest vast quantities of Web text.
In addition, Science Applications International Corporation (SAIC) received a $13 million contract to create a research framework for the development of reading system technologies and evaluate the performance of these technologies in support of DARPA’s program. The contract has a 5-year period of performance. SAIC will perform the work in Arlington, VA.
Additional Readings
- DARPA – Machine Reading Broad Agency Announcement [PDF]
- DARPA – Machine Reading: The Universal Text to Knowledge Engine [PDF]
- MaximumPC.com (June 30/09) – DARPA Developing Reading Machine To Record The Web
- Cnet.com (June 27/09) – Reading machine to snoop on Web
- GCN.com (June 26/09) – DARPA project seeks to teach machines better reading skills
- Wired.com (April 16/09) – Darpa Wants Brainy Machines to Replace Bored G.I.s
- DID (Jan 26/09) – NAVAIR’s Information Fusion Center. Provides an example of an advanced project that would make use of these capabilities.
- The Register (Nov 18/08) – DARPA seeks ‘Machine Reading’ AI auto-analysis bot