In its broadest sense, any repository of
information. A
filing cabinet full of text could be considered a knowledge base. However, usually "knowledge base" means a computer-accessible system of
knowledge (as opposed to mere facts), consisting of some set of
terms standing for
concepts, together with a heap of
relations among those terms that (hopefully) capture the
meaning of the terms.
Everything2 is one example of a knowledge base.
Cyc is another.
Corporate thesauri are also good examples of impoverished, but definitely machine-accessible, knowledge bases. Unlike Roget's Thesaurus (see Introduction to Roget's Thesaurus 1972), these thesauri organize words in a generalization hierarchy, and support crosslinks among words.