Vague Transformations in Information Retrieval

Information Retrieval (IR) is often confronted with different and heterogeneous collections of multilingual and multimedial content. Transformations are necessary in order to map between different representation schemes which are the result of different content analysis methods and different terminologies. Vague methods which have been sucessfully applied to IR seem especially suited to tackle the vague nature of such transformations. Statistical methods are reviewed and two experimental systems based on neural networks are presented.