ISR10 The Indexing Function chapter Joseph John Rocchio Harvard University Gerard Salton Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government. 2-6 inclusion relations are also of interest, e.[OCRerr]. cause-effect, process- products, etc. and these can be used to define additional transformations. For example, the class of'elements denoting processes can be identified and the. corresponding products listed. A document image containing a process term may then be modified to include the associated product term *and vice versa. In summary it is possible, then, to consider semantic index transformations which include a variety of term associations such that in principle a multiplicity 0£ index representations can be produced based on the same set 0£ machine recogr[OCRerr]izable li[OCRerr]istic features. The -. problem with such transformations, in [OCRerr]neral, is that a lar[OCRerr] number 0£ a .priori[OCRerr]semantic associations are possible among the index terms describing a [OCRerr]ven document. The correct associations are dependent on the context in which the terms are used so that a context free encoding such as is generally produced by machine processing does not necessarily improve the[OCRerr] accuracy of the index representation of information content. C. 3yntact''ic[OCRerr] Techniques In general both the statistical and semantic procedures discussed above i[OCRerr]ore the informati:Qn carried by the structural constraints'of the natural language. It is possible, however, to in' corporate a number of, syntactic reco[OCRerr]riition features into automatic indexing a'l[OCRerr]oritbms. One obvious use of this' [OCRerr]ind of information -is stem detection,' i.e.' recognition of the intrinsic association of the various'morphologica'l.forms of a'[OCRerr]ven word. Stem detection is readily