MONO91 Appendix B: Progress and Prospects in Mechanized Indexing appendix Mary Elizabeth Stevens National Bureau of Standards Assuming, however, that the input processing problems have been solved, we may ask what machines can do with respect to words in texts, or in portions of texts, that are avail- able in machine-useful form? The machines can ??read?? the words for purposes of shifting and sorting and can copy or reproduce the words in some desired order, as in a machine- prepared concordance. Machines can match input words with words already in store and thus exclude input words from further machine consideration (as by stoplists in KWIC (Keyword-in-Context) and other forms of derivative indexing) or stress certain input words with reference to a selective "inclusion" dictionary. Next, machines can tabulate and count, so that both absolute and relative word fre- quency data may be applied to either indexing or search-selection algorithms. Measure- ments of sequential distances between selected words in the input text may also be applied. Machine look-ups against a master vocabulary can provide automatic supplying of syndectic information, synonym reduction, lexical normalization, generic-specific subsumption, data with respect to previously observed word-word or word-subject co-occurrences. In addition, information can be provided as to the possible syntactic roles of input words. In the light of such machine capabilities, what can be said of the present state of the art in automatic indexing? Automatic indexing in the sense of machine-prepared indexes that are generated by the automatic extraction and manipulation of keywords, especially from titles, is of course widely used in KWJC indexes such as Chemical Titles and many others both in the United States and elsewhere. Fischer Z[OCRerr]/ provides a retrospective view of KWIC indexing concepts, including variants like KWOC (Keyword out of Context) and WADEX (Words and Authors Index to Applied Mechanics Review). She stresses the potentialities of linking such extraction indexing to selective dissemination systems and concludes: 11Plans for using the `Echo' satellites to link information centers around the world, in a world wide drive toward im- mediacy in information dispersion, will surely provide a place for KWIC indexes and for the KWIC concept. " Warheit Z3/ also reports that consideration is being given to combining selective dissemination systems and KWIC. Fundamental questions remain: How useful and how much used are KWIC and other machine-generated indexes based upon the extrac- tion of words from a limited portion of the author's own text? These questions relate to an important distinction between two quite different types of indexing. The distinction is that whereas "derived" indexing takes as index entries the author's own words in the title, the abstract or the full text, in "assignment" indexing an index term, descriptor, subject heading, or classification code is assigned to a document as an indicator of content and the term assigned does not need to be identical with any of the author's own words. We can report continuing progress in use of derivative indexing techniques such as KWJC, and also in experiments with automatic assignment indexing and automatic subject classification. Timeliness of index production is certainly one of the major virtues of KWIC. A similar timeliness is promised for automatic assignment indexing techniques provided that requirements can be kept sufficiently low with respect both to keystroking and computer processing. Intermediate results may be achieved by pre-editing, normalization, and post-editing techniques. Manual pre-editing to modify and supplement keywords in title, abstract, or portions of text has been used in permuted title and KWJC-type indexing from the punched card system that began operation in 1952 24/ to the `1notation-of-content" system developed for NASA 25/. Kreithen 26/ suggests a combination of derivative and assignment indexing, as follows: 226