A qualitative growth model for real world text knowledge bases

We introduce a knowledge-based approach to the analysis of real-world natural language texts, which addresses the particular needs of dealing with new knowledge items. In order to incrementally expand the underlying conceptual knowledge base, we exploit two kinds of evidence from the text understanding process, viz. qualitative knowledge about linguistic constructions in natural language texts and that about structural patterns in the emerging text knowledge base. These clues are used to generate concept hypotheses, rank them according to plausibility, and select the most credible ones for assimilation into the conceptual knowledge base. We demonstrate the feasibility of our approach and discuss the results of an empirical evaluation in terms of concept learning rates and learning accuracy.