MONO91 Other Potentially Related Research chapter Mary Elizabeth Stevens National Bureau of Standards Additional features include provision for the matrix of coefficients of association to change with time or with deliberate manipulation to improve performance. Thus: "Each normalized cell weight. . . rises and falls with time as each specific association increases or decreases in relative frequency. In this way, the matrix memory of associations changes with time, maintaining a cumulative pattern of associations reflecting one statistical characteristic of messages fed into it in the past... "In addition to this adaptive characteristic of changing memory with time and with changing inputs, the matrix is also readily subject to formal education. Any specific cell weight can be strengthened by repeatedly reading into the matrix memory the specific strings that contain the desired associations. For example, by introducing the strings is am, is are, am is, am are, and are am, we can 1/ 1-ncrease the statistical tendency of the tokens is, am, and are to be associated." - Experimental results have been obtained for a corpus of 500 bibliographic entries contained in DDC's Title Announcement Bulletin. In the case of a three-term query, 40 items were selected and ranked in probable relevance order, with selection based on a particular relevance score value threshold. The investigators then reviewed the abstracts of all 500 items and rated them as to relevance with respect to the query. Seven additional items were found, of which three would have been machine-selected with a le[OCRerr]s stringent selection threshold. For the remaining four, it is reported that they "were pcorly indexed and could have been judged not relevant by a human who depended upon the descriptor string only, as the matrix did, rather than upon review of the abstracts." 2/ 6.3 Clues to Index-Term Selection from Automatic Syntactic Analysis Several of the organizations and research teams most active in the investigation of linguistic data processing techniques, especially for automatic indexing, extracting and search renegotiation applications, are actively considering the use of clues derived from automatic syntactic analysis to improve criteria for machine selection of significant" words, phrases, and sentences from raw text. Such approaches, in general, however, are subject to the limitations of non-availability of sufficient corpora of text in machine-usable form, in the first place, and, even more importantly by the non- availability of satisfactory computer programs for complete syntactic analysis up to the 1/ 2/ Spiegel et al, 1963 [566], p. 17. Ibid, p. 34. 127