Controlled-term indexing (the method of choice for multimedia collections and still very popular for purely textual material), appears an expensive solution because it takes huge resources and manual indexing. It is not possible, however, to perform a well-founded asessment of various approaches to information retrieval. We discuss ways to improve controlled-term indexing and illustrate these by looking at the Condorcet project carried out at Twente by us and co-workers. We round o# with a discussion that, we hope, will raise more questions than it answers. Keywords: Knowledge-Based Systems, Language Technology, Information Retrieval, Multimedia, Information Extraction 1 Introduction Among the papers of the French mathematician and social philosopher Condorcet (1743--1794) found after his death, there is a proposal to assign each and every piece of knowledge a unique code. The code would serve two purposes: it could be used to organise libraries, and by comparing the codes with what ...