While being successful in providing keyword based access to web pages, commercial search portals, such as Google, Yahoo, AltaVista, and AOL, still lack the ability to answer questions expressed in a natural language. We explore the feasibility of a completely trainable approach to the automated question answering on the Web or large scale digital libraries. By using the inherent redundancy of large scale collections, each candidate answer found by the system is triangulated (confirmed or disconfirmed) against other possible answers. Since our approach is entirely self-learning and does not involve any linguistic resources it can be easily implemented within digital libraries or Web search portals.