ISR10 Appendix A: The Smart System appendix 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. A-3 Document index images enerated by this process may be subjected to a number of additional modifications. A variety of transformations based on a pre-specified hierarchical structuring of the elements of the index language is provided. Alternatively, relations among index terms derived from statistical associations in a given collection may be used for modifying index images. The system, therefore, may provide a variety of representations for input documents based on the initial dictionary lockup and subsequent transformation rules defined on the index language. ([OCRerr]ote that the index images used experimentally for this thesis were generated by a lookup using version 2 of the SMAI[OCRerr]T thesaurus with no. phrase detection and no additional semantic transformations.) [OCRerr]. Search Thequest Formulation Search requests in the SNART system are introduced.directly in the natural language and may be treated exactly as are document texts. Requests, therefore, may be subjected to all or any subset the content analysis procedures available for document processing. In addition to varying the index image of a search request by the sequence of analysis procedures to which it is subjected, a number additional query modification procedures4 (including the relevance feedback technique discussed in[OCRerr] chapter [OCRerr]) are being considered for inclusion into the system. of of C. Q[OCRerr]ery-Document Matching The flexibility provided by the computer allows the S[OCRerr]ART