The work herein reports on research conducted in the domain of Ontology Learning from text collections, without background or prior knowledge. The research focuses on the automated construction of a subsumption hierarchy from text corpora using machine learning methods. Within this work, we define methods to identify concepts in large document collections and arrange them in a subsumption hierarchy. We focus on Bayesian methods and we focus on the non-parametric nature of these methods, in the sense of automatically determining the size and structure of the learned hierarchy. An evaluation method is also proposed that supports the automated evaluation of ontology learning, by means of comparing learned ontologies with gold standard ones. We evaluate the task of ontology learning using the proposed gold-standard evaluation method, that goes beyond lexical similarity between ontology elements. Finally, we define a set of evaluation measures that penalize the learned ontology according to its lexical and structural differences from the gold standard ontology. For each learning method, as well as for the evaluation method, we present experimental results on real datasets that indicate the potential of the proposed methods.