Drawing upon the extant literature, we propose a classification schema of ontology learning techniques, as shown in '그림 3'. The top-level consists of three categories: statistics-based, rule-based, and hybrid techniques. The majority of ontology learning techniques are unsupervised because training data annotated with ontological knowledge are commonly not available. As a result, statistical techniques have been often applied in ontology learning. A statistical model is typically represented as a probabilistic network that indicates the probabilistic dependencies between random variables. The statistical information computed from observed frequencies or joint distributions of the terms is used to determine concepts and their relations. Different approaches vary in how this probabilistic network is generated and which method is applied to combine individual distributions. Maximum Likelihood Estimation and Bayesian approaches are typical examples. Rule-based approaches require matching to pre-defined r es or heuristic patterns in order to extract terms and relations. A rule-based model is typically represented as a set of rules consisting of condition testing and action execution, such as dependency relation analysis and anaphoric resolution. Hybrid approaches leverage the strengths of both statistics-based and rulebased approaches.