An Empirical Evaluation of a System for Text Knowledge Acquisition

We introduce a formal model and a corresponding system architecture for the acquisition of new concepts from real-world natural language texts. Our approach is centered around the linguistic and conceptual \quality" of various forms of evidence underlying the generation and reenement of concept hypotheses. Based on a ter-minological (meta)reasoning platform, hypotheses are continuously annotated by a stream of linguistic and conceptual evidence, preferentially ranked and, nally, selected according to their overall credibility. We discuss the results of an empirical evaluation study, concentrating on the system's learning rate and learning accuracy. Abstract. We introduce a formal model and a corresponding system architecture for the acquisition of new concepts from real-world natural language texts. Our approach is centered around the linguistic and conceptual \quality" of various forms of evidence underlying the generation and reenement of concept hypotheses. Based on a terminological (meta)reasoning platform, hypotheses are continuously annotated by a stream of linguistic and conceptual evidence, preferentially ranked and, nally, selected according to their overall credibility. We discuss the results of an empirical evaluation study, concentrating on the system's learning rate and learning accuracy.