The system described in this paper is claimed by its authors to “use production rules from artificial intelligence to define a hierarchy of retrieval subtopics, with fuzzy context expressions and specific word phrases at the bottom.” Older practitioners of information retrieval might prefer to characterize RUBRIC as a thesaurus-based system with term weights, since the rules do nothing more than provide a means for encoding a definition hierarchy, supplemented with weights. Recasting the older approach into the style of production-rule AI systems adds nothing essential. It may of course be more convenient to use the AI style, making it easier to add new features or to do system debugging. The authors appear to have made that assumption, but I suspect others might disagree.
As to the merits of the system itself as a retrieval device, very little can be said because it had hardly been tested as of the time of writing. The only experiments reported were done on a collection of 30 news service stories. While such databases are useful for debugging a system, essentially nothing can be inferred as to system performance in an operational environment.
In short, the system reported here may have merit, but it is too early to make any judgment. Researchers in information retrieval might be interested in seeing how a currently fashionable software building paradigm can be used in their discipline.