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-55 the examples given show that the thesaurus and the hierarchy are often successful because of the precision device effect achieved by the weighting scheme. Confirmation of the fact that this phenomenon is largely responsible for the improvements gained by thesaurus in the IRE-3 and ADI collections is obtained from the data of Fig. 32, where performance results from Figs. 5 and 6 are represented with, and without, the weighting process to show how the thesaurus offers greater improvement over stem when the weighting scheme is in usee The Cran-l collection does not in this instance show this result, probably because of the effect on the cosine correlation of the change in weighting; alter- natively, it may be explained as yet another instance of the difference in behavior between the Cran-l and the others. 8. Further Studies Required Since the conclusions of this section have already been stated in part 6, this final part enumerates some topic areas for further in- vestigation that may be directly or indirectly suggested by the preceding analysis. Eleven studies are listed: 1. The effectiveness of all five of the dictionary con- struction rules must be established by the construction of a series of versions of a given dictionary, so that the relative importance of rules about word frequency versus rules about synonymy can be established. As a start in this direction, a second version of the ADI semi-automatic thesaurus is under test.