IRS13 An Analysis of the Documentation Requests 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. x-8 than single words. These examples of difficult requests point out two areas in which future work is required. The first is the problem of enibiguity caused by natural language, which may partially be handled by sophisticated recognition procedures (to include negative statements, for example), but may in other cases only be handled by introducing constraints to the free statement of the request. The second problem is that of making synonym connections in cases where a generic term is used. A possible solution is the provision of more than one synonym dictionary, which would include one containing some quite large groupings of many words into few concepts to handle the difficult cases. [OCRerr]. Relevance Decisions Since both request preparers were alone responsible for the requests and relevance decisions produced, there arose no possibility for disagreement during the setting up of the test. Evidence suggests that a consistent and conscientious job was done, although it is very easy to argue that the judges were not competent, or that real information needs did not arise, and so on. Measurement of the accuracy with which this artificial procedure can simiulate real user requests, needs, and relevance decision awaits a care- fully controlled comparative test. To submit the actual judgments made to a panel of judges for their opinion would undoubtedly reveal disagreements with the request preparers, and probably among the judges themselves; such a procedure would then serve no real purpose at the present time. A cursory look at the decisions has been taken, and some discrepancies are noted in order to illustrate the probable types of deficiency that may exist.