The recent considerable growth in the amount of easily available on-line text has brought to the foreground the need for large-scale natural language processing tools for text data mining. In this paper we address the problem of organizing documents into meaningful groups according to their content and to visualize a text collection, providing an overview of the range of documents and of their relationships, so that they can be browsed more easily. We use Self-Organizing Maps (SOMs) (Kohonen 1984). Great efficiency challenges arise in creating these maps. We study linguistically-motivated ways of reducing the representation of a document to increase efficiency and ways to disambiguate the words in the documents.