MONO91 Operational Considerations chapter Mary Elizabeth Stevens National Bureau of Standards assignment they had co-occurred)3 the machine had a sufficient basis in the input material for the derivation of a selection-score for at least lZ descriptors for each new item. The items were closely similar to, though not identical with, the source items from which the word associations with descriptors assigned had been drawn. The sample is obviously critically small. Nevertheless, the possibility that extensive clue word lists, notwith- standing the incorporation of trivial and even erroneous associations, can be used as effectively as smaller, more precise, and more carefully tailored lists, but with signifi- cant gains in memory space or computational reqilirements, is suggestive. A somewhat related conclusion, again reflecting the effect of processing requirements, is stated by Needham as follows: 11The main point to be made is that theoretical elegance must be sacrificed to com- putational possibility: there is no merit in a classification program which can only be applied to a couple of hundred objects." 1/ In KWIC type derivative indexing by machine, except in terms of allowable character sets and word-lengths conveniently processed, the problem of appropriate programming languages does not arise to any serious extent. For the processing of material in research on natural language text, however, the choice of interpretative and compiler types of auto- matic programming languages may involve computational requirements which, while being inappropriate in a production situation, offer considerable flexibility and versatility for experimental purposes. Examples of special programs of this type include the use of Yngve's COMIT by Baxendale and Knowlton, the development and use of FEAT by Olney, Doyle, and others at SDC, and the use of list-processing techniques in the General Inquirer system. z/ Yngve describes the use of his program as follows: "COMIT has also been used in the experimental work in information retrieval of Baxendale and Knowlton at IBM. The purpose of their COMIT program was to accept as input the title of a document and to produce as output, not only descriptors, but pairs of descriptors which are roughly of the form adjective-noun. The purpose of the work is to automatically generate, from document titles, retrieval words of a more specific nature than simply Boolean functions of the existence of certain words in a title. " 3/ The FEAT program was designed originally for word and significant-word-pair frequency counts. Olney describes the program in part, as follows: "FEAT is designed to perform frequency and summary counts of words and word pairs occurring in its natural text input; i.e., text written it' ordinary English and transcribed into Hollerith code according to some set of keypunching rules. To focus attention on the semantic aspects of word pairs rather than on their syntactic aspect, pairs of which one member is a function word, such as `the', `is', `by', etc., are excluded. "Using a bucket list structure of the type proposed by C. J. Sheen in FN-1634, the program sorts each incoming word serially, constructing a list within each of Z56 buckets for good words of a given alphabetic range ... and another list within each good word entry for the Doubles and Reverses which will be ordered alphabetically 1/ Needham, 1963 [433], p. 8. Stone, et al, various references, p. 137 of this report. 3/ Yngve, 196Z [655], p. Z6. 170