A method and apparatus for discovering polysemous words and classifying polysemous words found in web documents. All document corpi in any natural language have words that have multiple usage contexts or words that have multiple meanings. Semantic analysis is not feasible for classifying all word occurrences in all documents on the web, which contain trillions of words in total. In addition, semantic analysis typically cannot distinguish multiple usages of a given meaning of a given word. In one embodiment of this invention, polysemous words in natural languages can be discovered by analyzing the co-occurrence of other words with the polysemous word in web documents. In one embodiment, the multiple meanings and usages of a polysemous word can be determined by analyzing the co-occurrences of other words with the polysemous word. No semantic analysis is used in discovering or classifying polysemous words.