DARPA project seeks to teach machines better reading skills
The Defense Advanced Research Projects Agency is giving new meaning to the term “machine readable” with a program to develop an automated reading system that would close the gap between natural text and artificial intelligence systems.
With funding from DARPA, the Air Force Research Laboratory has awarded a $29.7 million contract to BBN Technologies to develop the system under DARPA’s Machine Reading Program.
The five-year plan calls for BBN, which has a lot of experience with natural language processing, to develop a universal text engine that would convert text into representations understood by artificial intelligence systems, the company said. The challenge includes developing a system that works across diverse linguistic structures and axioms and can function on the scale of the Web.
The system would be used initially for military intelligence analysis, according to BBN, but it could be expanded into other applications, such as analysis of digitized content in libraries.
“Such a system could eliminate many of the impediments to stability that our military faces, such as a lack of understanding of local customs, and give us the ability to assess global technology developments continuously,” said Prem Natarajan, BBN’s vice president of speech and language processing.
Kevin McCaney is the executive editor of GCN. Follow him on Twitter: @KevinMcCaney.