Charting Paragraph Semantics
USAF SBIR Phase I Proposal on Topic AF06-060
By Dan Corwin, Lexikos Corporation, January 11, 2006
Lexikos has devised MODELER software to semantically represent the topics of English phrases and sentences. Run as the back end to any English parser, it uses a Topic Map to chart each input paragraph’s semantic content, transcribed into metadata. Charts expose the predicate-argument structures for clauses and relations, and obey constraints defined in user-supplied lexicons.
A desktop client functioning as an interactive spelling corrector helps the user convert paragraph text into charts, soliciting human assistance as required. This form-driven process can “train” MODELER on new vocabulary in the user’s domain of discourse. Significant bulk lexicons and ontology models for each domain are required as a base to support such incremental additions.
In Phase I, we will demonstrate that charting is feasible by passing MODELER limited-grammar text inputs in the domain of HL7, a standard ontology widely used to document the intentional healthcare acts taken or ordered by or reported to professional caregivers or their organizations.
In Phase II, Lexikos will produce a well-trained Java web application ready for deployment. It will let healthcare actions be discussed more naturally, yet still transcribe to useful HL7 metadata, and illustrate information-extraction methods that can generalize to cover many other domains.
Potential Commercial Applications
Our national power to exploit intelligent agents is impeded by our inability to communicate with them. Charting software can ease such limits, and greatly enhance voice analysis, language translation, multimedia indexing and retrieval, message routing, content management, knowledge representation, intelligent computer-based training and related areas of modern data processing.
Our charts can be accessed under both Topic Map and RDF metadata paradigms. This combination will stimulate new R&D in several related technical communities, foster new national progress in knowledge-based applications, and enhance the prospects of long term success for all Lexikos text analysis products.