Ontologies are designed to facilitate knowledge sharing and reuse between humans and software agents. In terms of the level of abstraction, ontology representation languages form a spectrum ranging from natural language at one end to formal language at the other. The two extreme ends best satisfy the needs of humans and software agents respectively. In order to bridge the gap between them, we may allow humans to access ontologies using natural language by mapping natural language words/phrases to low-level ontology concepts automatically while keeping the toplevel ontology abstract.