ISR11 Operating Instructions for the SMART Text Processing and Document Retrieval System chapter 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. 11-3 The analysis of English language text for the detection of concepts is performed without human intervention. Text is analyzed by a set of programs which operate in conjunction with two types of inputs: a) A set of dictionaries, grammars, and hierarchies specifying the relations between properties of the input English text and of the concepts. The simplest possible dictionary, for ex[OCRerr][OCRerr]le, associates with each English word a distinct concept. A more sophisticated dictionary, consisting in fact of a word thesaurus, will define a many-to-many mapping of English word stems into the concept vector space in which some words are isolated as c[OCRerr]nmon words, some are considered anbiguous and resolved into their various possible meanings, and some are combined with syno[OCRerr]ous terms. b) A set of specifications which describe in what way the various content analysis programs are to be applied and which dictionaries are to be used. Specifications are used by the program to decide what documents are to be processed, what dictionaries should be used, what algorithms for associating concepts with documents should be employed, how much weight each contribution to the document representation should be given, what procedures should be utilized to compare requests with documents, and what output is desired from the run. Once the text analysis has been performed, and concept vectors are available for all documents and all requests that have entered into the system, the programs proceed by comparing the requests with the documents to determine which documents are to be identified as `tanswers't to the requests. This list may then be compared with a 1tcorrect" list, obtained as a result of a manual operation and supplied by the programmer, and various evaluation measures may be computed automatically.