SBIR-STTR Award

Lexikos has prototyped common lisp software able to systematically find, model, and relate the central topics of well-formed English text, then record this data in a modular way that lets other software make use of it. The foundation of our approach is a knowledge representation formalism able to encode the semantics of English word meanings and phrases. It blends frames and rules under a message passing scheme, under conventions directly reflecting the apparently ad hoc nature of English phrase structure rules. The linguistic conventions used in our formalism make it novel, general, powerful, and elegant. With it, anything described in simple English can be mapped into isomorphic lisp data and readily incorporated into a hierarchical, network-like data base, even on small host machines in the IBM PC class. This kind of data base will directly meet the goals of this SBIR topic. We propose to describe our formalism in depth, explain our implementation of it, give concrete examples of its current applications, and define the work still needed to refine it into production quality utility software.

In Phase I work on this topic, Lexikos demonstrated prototype "transcriber" software. With operator aid, it can isolate, model, and relate the central topics of well-formed English text, then output this data into a text file written in a formal modeling language called Inglish ("idealized English"). Inglish outputs from the transcriber give a detailed linguistic model of the input text and accurately depict meanings for its individual words and phrases. When they are used as a data base or knowledge base by external applications software, the topics and surface assertions of the original text become "understandable" to that application software, and they can be used to help intelligently guide its own separate activities and decision-making. For Phase II, Lexikos proposes to expand our current prototype transcriber into a full scale beta-site version, organized to help the air force model, query index, correlate, summarize, reorganize, and redistribute bulk textual data found within the class of document known as technical orders.