Introduction As the on-line world grows and increases its role in our daily lives, the problems of searching, categorizing, and understanding textual information become ever more important. While researchers and practitioners have made much progress in these areas over the last thirty years, anyone who has gone to the World Wide Web seeking information and returned with more frustration than answers can attest that much work remains. Today's important issues cover topics such as scalability, user interfaces, and techniques that exploit the unique hypermedia features of Web environments. To support our research in these areas, the Text Analysis and Advanced Search department at the IBM T. J. Watson Research Center has developed an experimental probabilistic text retrieval system called Guru[4]. Guru was originally built to explore new probabilistic ranking algorithms, and now serves as a test-bed for much of our text analysis, search, and categorization work. Guru may be run as