ISR11 Information Analysis and Dictionary Construction chapter G. Salton M. E. Lesk Harvard University Gerard Salton Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government. IV- 30 below their parents and further to the right. The hierarchy of Ffig. 3 also provides for the inclusion of cross references from one concept to another, which are coi'[OCRerr]ected to the original concept by broken lines. Such cross references represent general, unspeci- fied types of relations between the corresponding concepts, and receive in general a different interpretation than the generic inclusion relations normaliy represented by the hierarchy. It would be nice if it were possible to give some generally applicable aigorith[OCRerr] for constructing hierarchical zub[OCRerr]ect arrangements. This is, in fact, a topic which has preoccupied many people including mathematicians, philosophers, and librarians for many years. In ge[OCRerr]eral, one can sa[OCRerr] that broad concepts should be near the top of tree, whereas specific concepts should be near the bottom; furthermore there appears to be some relation- ship between the frequency of occurrence of a given concept in a document collection, and its place in the hierarchy. More specifically those concepts which exhibit the highest frequency of occurrence in a given document collection, and which by this very fact appear to be reasonably common, should be placed on a hi-her level than other concepts whose frequency of occurrence is lower. Concerning the specific place of a given concept within the hierarchy, this should be r[OCRerr]de to depend on the user population and on the type of expansion which is most often requested. Thus, a concept corresponding to 1tsyntactic dependency tree1 would most reasonably a[OCRerr][OCRerr]ear under the broader category of 11syntax1t, which in turn could appear under the general class of ttlanguage't, assuming that the user population consists of linguists or grammarians; on the other hand, if the users were to be mathematicians or algebraists, then the t1syntactic dependency trees[OCRerr] should probably appear