IRS13 Thesaurus, Phrase and Hierarchy Dictionaries chapter E. M. Keen Harvard University Gerard Salton Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government. VII-46 in aerodynamics. This does not rule out the factor of quality of the thesaurusf but since the thesaurus-3 was constructed with the thesaurus rules by the same person who made the IRE thesaurus-3 and ADI thesaurus-l, it does not seem likely that the Cran thesaurus-3 is really bad. A further reason for accepting the Cran-l environment as being responsible lies in the findings of the Cranfield Project [17] in which the "quasi- synonym" index language was found to be inferior to the 11word form31 (i.e. stem) language. The quasi-synonym list was rearranged in certain respects to become the SMART Cran thesaurus-l, and this dictionary does perform worse than stem for both abstracts and indexing. The reason for this result offered by the Cranfield Project was that imprecise terminology is not helped by a dictionary grouping of the thesaurus type since any help given to the poorly matched relevant documents is offset by an increased number of retrieved non-relevant documents. This conclusion needs modification in two ways, in the light of the SMART results. First, terminology alone is unlikely to be the only explanation, since the ADI collection on documentation is believed to use terminology at least as imprecise as a[OCRerr]rodynamics, and the thesaurus-l does provide some advantage for that collection. Second, some amount of grouping of aerodynamics terms does give a slight performance increase, since the thesaurus-3 dictionary gives better results than stem or suffix `5' over some parts of the curve. Another point on which the Cranfield con- clusion may not apply to SMART is in the use by SMART of a weighting scheme of the type not tested at Cranfield. In fact, without the weighting scheme in use, the ADI text result does follow the Cranfield