MONO91 Other Potentially Related Research chapter Mary Elizabeth Stevens National Bureau of Standards The study of the position of keywords in the text and the syntactical relationship which exists among them will show the way to automatic ab- stracting and the use of more sophisticated retrieval Systems." 1/ Plath suggests that, given a computer program to perform the parsing and syntactic diagramming of a text sentence, the results can serve quite usefully to augment the selection criteria based initially on statistical techniques, such as word-frequency counting. He says, for example: "Another possible application of the outputs of the sentence diagramming program is their employment as an aid in language data processing for purposes of information retrieval, particularly in systems for automatic literature abstracting of the sort proposed by Luhn (1958). The feature of the tree diagrams which is pertinent here is that the main components of a clause, including subject, verb and object, always correspond to the `main topics' in an outline, and are therefore located at the upper levels of the tree. When the words on these upper levels are considered apart from the lower-level structures which modify them, they often summarize the content of the sentence in a sort of `newspaper headline' or `tele- graphic style'." [OCRerr]I The problems of multi-level selection, or screening, such that machine programs for selection of the most probably significant words, phrases, or sentences can be focussed upon the most probably content-relevatory areas of text, are treated here, as also by Salton, in the sense of a cutting-off at a given depth in the analyzed syntactic structure. A potentially important contribution to the future prospects for automatic indexing, however, lies in the "discourse analysis" and "transformational linguistics" approach of Harris (1959 [254]), where condensations and concentrations of similarities and differences of topical interest may hopefully be achieved. Harris himself suggested, at least as early as 1958, applications of his approach to both automatic indexing and abstracting. A goal of the analyses he has proposed is to identify `kernels' of linguistic expression, having first, by various transformations such as from passive to active voice, brought together different ways of saying the same thing. He then suggests not only machine operations to normalize by application of his trans- formational rules but also to determine: Which kernels have the same centers in different relations (e.g. , with different adjuncts), and other characterizing conditions. The results of this comparison would indicate whether a kernel is to be rejected or transformed into a section... of an adjoining kernel, or stored, and whether it is to be indexed, and perhaps whether it is to be included in the abstract." 4/ 1/ 2/ 3/ 4/ Levery, 1963 [359], p. 236. Plath, 1962[474], pp.189-190. See also Thorne, 1962 [605], p. v: "The approach followed requires that the com- puter itself syntactically analyse input text in order to convert it into special form called FLEX, which preserves only that syntactic information which is useful for data retrieval purposes." Harris, 1958[254], p.949. 130