DIMACS TR: 2000-29

Technique of semantic headers: a manual for knowledge engineers

Authors: Boris Galitsky


This manual addresses the issues of knowledge representation in the form of semantic headers (SH) for conversion of a textual document into a form, appropriate for question answering. A set of semantic headers is intended to formally represent the essential idea of a document with respect to possible questions, such that this document would serve as an answer. The knowledge base then includes the textual answers, structured by means of assigned semantic headers. The formal representation of the natural language (NL) query is then matched against the knowledge base in the form of a set of these headers. The logical mechanisms of matching query translations against semantic headers and derivation of semantic headers themselves is presented, as well as applications of meta-reasoning, default reasoning, reasoning about action and time, and graph representation of answer classification.

This technique shows superior performance over the knowledge systems based on syntactic matching of NL queries with the prior prepared NL representation of canonical queries, and the knowledge systems based on the fully formalized knowledge. Our approach gives the higher precision of answers than the former one because it involves the semantic information in higher degree. At the same time, in the logically complex and poorly structured domains SH technique gives more complete answers, possesses higher consistency to context deviation and is more efficient than the latter approach because the full knowledge formalization is not required.

The manual of semantic header technique is intended to assist the linguists and knowledge engineers in the creation of question answering systems for the sophisticated vertical domains. The manifold of coding samples and extended discussions help to deeper understand the peculiarities of using logic programming in the knowledge representation. The tutorial part is followed by the qualification test using the multiple-choice questions.

Paper Available at: ftp://dimacs.rutgers.edu/pub/dimacs/TechnicalReports/TechReports/2000/2000-29.ps.gz

DIMACS Home Page