Text condensation as knowledge base abstraction

A model of knowledge-based text condensation is presented which has been implemented as part of the text information system TOPIC. Two major processes, text parsing and text condensation, are considered in detail, with emphasis on the latter. Based on principles of semantic parsing, the text parser serves the purpose of augmenting the initial domain knowledge base with the knowledge encoded in a text, thus generating a specific text knowledge base. A condensation process then transforms these text representation structures into a more abstract thematic description of what the text is about, filtering out irrelevant knowledge structures and preserving only the most significant concepts. Finally, a hierarchical representation of the thematic units of the text is generated in terms of a text graph which supports variable degrees of abstraction for text summarization as well as content-oriented retrieval of text knowledge.>