MONO91 Indexes Generated by Machine-Automatic Derivative Indexing chapter Mary Elizabeth Stevens National Bureau of Standards at Rutgers in indexing of a book by computer programs (1963 [zo] and [22]) is an example of such modified derivative indexing. Specifically, Artandi's method involves: (1) Establishment of a list of key terms appropriate to a given subject area to be used as an inclusion list for word extractions from text. (2) Application of an appropriate syndetic apparatus to be used in the compilation and ordering of the inde[OCRerr] entries. (3) Means for the automatic selection of index entries other than those on the pre-specified inclusion list, especially for the selection of proper names. The text used by Artandi for her study consisted of a 59-page chapter on halogens from J. W. Mellor1s Modern Inorganic Chemistry. This text was keypunched with special tags being assigned to indicate the page numbers and the incidence of capitalized words in the text. Text words greater than three characters in length were first checked against the inclusion dictionary of `1detection terms". There was, in addition, an "expression term" dictionary which constituted the vocabulary of the final index and in which a given expression term might or might not be identical with the corresponding detection term. Cross-references were supplied by a program routine which checks the index term list against a list of expression terms with their detection terms grouped under them and which compiles cross-reference entries, one for each detection term associated with an expression term appearing on the index list. For her experimental corpus, Artandi's program developed 363 page references, 138 different index entries and 35 cross-references. She compared these results with those obtainable by conventional human indexing with respect to the factors of heading density (ratio of number of entries to number of words in the book), entry density (ratio of the number of page references to the number of pages), and distribution (ratios of entries for chemical compounds, proper names, and subject entries to the total number of entries. No indexing errors were found in the computer-generated index for a 5 percent random sample of the pages of the corpus, but five Omissions were found in the machine indexing of these sample pages. Artandi concluded, however, that although the quality of indexing appeared favorable, the costs, which approximated $1.50 per page indexed, were impractically high. Book indexing by computer has also been investigated by Maloney, Dukes, and Green at the Army Biological Laboratories, Fort Detrick, 1/ Maryland.[OCRerr] Input is based on the by- product paper tape generated when the manuscript is typed on a tape typewriter. The paper tape is in turn converted.to punched cards which are then processed by a UNIVAC SS-90 II computer in an editing run that deletes unrecognizable codes and then stores page, 1/ C. J. Maloney, private communication. A report by C. J. Maloney, J. Dukes, and S. Green, "Indexing reports by computer" is in process of preparation for publication. 72