MONO91 Indexes Generated by Machine-Automatic Derivative Indexing chapter Mary Elizabeth Stevens National Bureau of Standards To some extent, however, the use of human editors to improve the product of KWIC type indexing defeats the initial purpose of a quick and purely clerical or mechanical process. Thus, Dowell and Marshall argue: "...The basic permuted-title index can be substantially improved by editing and re- writing the titles before they are submitted to the computer. . . . But this of course, destroys the great advantage claimed for the permuted title index, `that it is a purely clerical process'. Intellectual effort has entered the picture again and we are back where we started.' 1/ In the extreme case, the re-introduction of intellectual effort is in effect the re-introduc- tion of conventional human indexing, with the machine's role limited to that of compilation, as in the case of the "notation-of-content" statements prepared for NASA's STAR System (Slamecka and Zunde, 1963 [561]; Newbaker and Savage, 1963 [430]). Kennedy suggests instead, therefore, that the augmentation might be accomplished by the authors themselves. However, it may then be pointed out, as by Bernier and Crane, for example, that the supplementation of titles before publication in order to provide suitable additional indexing words would be "awkward, space[OCRerr]consuming and difficult". They continue: "It would call for the attention of index experts at the manuscript stage, which would delay publication and expand the total indexing effort. Furthermore, good, thorough indexes are based on the full information of abstracts and papers, not on their titles only." 2/ An alternative method for title augmentation to improve the quality of KWIC indexing is therefore to establish procedures for machine selection of significant words from more of the text than just the titles alone. In fact, Luhn himself did not limit his technique as originally proposed to titles only but indicated that the process could be performed at various levels: title, abstract, or full text. 3/ In the 1958 permuted index to the ICSI preprints, entries were derived from titles, author's names, author affiliations, headings withinthe paper, figure and table captions, and sentences and phrases taken directly from text. [OCRerr]4I Combinations of human and machine procedures based on sentences and phrases selected from text are described by Herner who cites a two-fold advantage: "First, it is not wholly dependent on the informativeness or lack of informativeness of titles and bibliographic citations, and, second, it affords a greater depth of analysis than is generally possible where titles or bibliographic descriptions alone are used." 5/ 1/ 21 3' 4/ 5/ Dowell and Marshall, 1962 [159], p. 324-325. Bernier and Crane, 1962 [56], p.117. Luhn 1959 [381], p. 289. Citron, et al, 1958 [120], p. i. Herner, 1963[264], pp. 1-2. 69