Tell Me About
Navy SBIR 2012.2 - Topic N122-136
ONR - Ms. Tracy Frost - email@example.com
Opens: May 24, 2012 - Closes: June 27, 2012
N122-136 TITLE: Tell Me About
TECHNOLOGY AREAS: Information Systems, Human Systems
ACQUISITION PROGRAM: PM Intelligence Systems Acquisition Program, APTS (Mobile ISR-C2) FNC
OBJECTIVE: Develop and demonstrate a capability to automatically respond to "tell me about" questions with finished information products.
DESCRIPTION: Military Intelligence and Operations staff spend large amounts of time preparing reports and briefings by searching for related information from many data stores, estimating data relevance, and then fusing and formatting relevant data as information products. Active Wiki technology has been used to update web pages about specific topics, but these sites require advance knowledge of where to find relevant content. Additionally, search engines based on user input of words to find various types of data are common.
This SBIR topic will advance the state of the art by providing a capability to accept broader "tell me about" questions (e.g. persons, groups, places, events) and by requiring the system output to look like a finished information product. To accomplish this, a system is needed that can semantically understand the content requirement of questions and semantically enrich raw unstructured data. Progress made in automated semantic extraction, or the extraction of entities with information/context associated with that entity could be applied to this problem. The recognition of content, in the form of word frames or themes is also applicable to the topicís challenge. Due to these and other advances in semantic natural language processing, it should now be possible to translate the information content requirements of a "tell me about a person, event or place" question to machine understandable ontology. The Phase I performer will need to select several "tell me about" questions to consider, translate these to content models, automatically discover, fuse, and import relevant information, calculate how much of the content required was found, and package discovered/distilled data as a finished information product.
Key technical challenges include: question content modeling, semantic enrichment of unstructured text, automated tracking of the sufficiency of discovered content to a semantic question, and automated production of a finished product in a commonly used format (e.g. output to an MS Office application). The Navy is interested in innovative R&D that involves some measure of technical risk. Proposed work shall have technical and scientific merit. Creative solutions are desired.
PHASE I: Select several "tell me about" questions to consider. Reduce the technical risk associated with the development of a system that can model the content required to address a complex question, semantically enrich raw data, and produce a finished intelligence product relevant to the question asked. Track key technical performance parameters. Conduct a demonstration at the end of the Phase I effort that clearly shows how much risk, relative to the production of a full prototype system, has been mitigated.
PHASE II: Develop and demonstrate a prototype system that is capable of producing finished intelligence products from "tell me about persons, groups, places, events, etc." questions by discovering the full answer to these questions from semantically enriched unstructured text. A proof-of-concept should be shown with relevant document corpus and content completeness of 80% and content inclusion accuracy of 90%.
PHASE III: Produce a system capable of deployment and operational evaluation. The system should address "tell me about" questions that have relevance to intelligence fusion cells. Increase performance over Phase II demonstration that can produce finished products with a completeness of 80% and accuracy of 90%.
PRIVATE SECTOR COMMERCIAL POTENTIAL/DUAL-USE APPLICATIONS: There are many potential commercial applications, including law enforcement and internet search engines, for an application that can produce a summary word document vice a list of links in response to a question about a person, or any other topic.
2. The Stanford Natural Language Processing Group. 2011. "Shallow Semantic Parsing." Accessed December 2. http://nlp.stanford.edu/projects/shallow-parsing.shtml.
3. Peñas, Anselmo and Eduard Hovy, 2010. "Semantic Enrichment of Text with Background Knowledge", Proceedings of the NAACL HLT 2010 First International Workshop on Formalisms and Methodology for Learning by Reading. W10-0903. Accessed December 2, 2011. http://aclweb.org/anthology/W/W10/W10-0903.pdf.
KEYWORDS: Natural language processing; semantics; content modeling; auto-generation; complex queries; meta data tagging; machine understanding