Knowledge Triage for Military Leadership

Period of Performance: 02/02/2004 - 08/02/2005

$100K

Phase 1 SBIR

Recipient Firm

Knowledge Analysis Technologies, LLC
4940 Pearl East Circle, Suite 200
Boulder, CO 80301
Principal Investigator

Research Topics

Abstract

Intelligence for combating terrorism is made difficult by the hundreds of minor languages and dialects potentially spoken by terrorists. There is a pressing need for automated methods that greatly reduce reliance on scarce bilingual analysts and automatic translation systems. The project objective is to develop a computer-based tool that will allow military leaders and intelligence analysts to quickly and accurately select only important documents for translation and/or analysis. An important aspect of the problem is known as Knowledge Triage. For example, in Iraq, over one million pages of paper documents, the majority in poorly handwritten Arabic script, have been collected, Most of this is almost certainly of no intelligence value, while some may contains hints, and a very small number "smoking gun" information. An existing cross language retrieval technique based on Latent Semantic Analysis (LSA) is capable of performing the triage, but only if given text in electronic form. In the proposed system a target language speaker would read paper documents into a speech recognizer, and the system would assess their possible relevance and summarizes their contents in English keywords. The system is robust to speech recognition errors and can be adapted to a new language in a few weeks.