MONO91 Automatic Assignment Indexing Techniques chapter Mary Elizabeth Stevens National Bureau of Standards None test item bodies of of the experiments has so far encompassed testing of anything but very small samples and the dangers of extrapolating from so small and so specialized data should be clearly recognized. Mooers identifies these dangers in terms of "The Silent Postulate: (real people) That (real documents) can somehow (real jobs to do) be eliminated from the experimental study, and that (role-playing people) (substitute documents) (imaginery jobs) 1/ can be substituted and still give valid experimental results." - In most of the experiments in automatic indexing conducted to date, indexing and classification schedules have been especially designed, or evaluations made, specifically for the purposes of these tests. Williams, however, stresses the point that the material used in his experiments had been "classified by professional indexers for the purposes of actual retrieval." 2/ A similar claim can be made for SADSAGT, as noted by Mooers. 3/ Swanson's news item work also obviously relates to real items and implies a real job to be done, but is directed, as noted, to a class of material not generally comparable to that found in documentation operations on scientific and technical literature. In contrast with the treatment of each document as a self-contained entity without reference to any other documents, as is the case for derivative indexing, all of the automatic assignment indexing experiments, by virtue of the fact that they are assign- ment techniques, do to some extent embody the effects of a consensus of a particular collection, or a consensus of prior indexing, or a consensus of human subject content analysis applied to sample documents, or some combination of these effects. The SAD- SACT method, in addition, wherever cited titles are available for new items, takes advantage of terminology other than the author's own as a source of clue words. Other proposed methods of assignment indexing, such as the use by Salton, Lesk, and Storm of citation-pattern similarity data, would carry the latter principle even further. 1/ 2/ 3/ Mooers, 1963[424], p.5. Williams, 1963 [642], p. 162. Ibid, p. 5. 04