CRANV1P1 ASLIB Cranfield Research Project: Factors Determining the Performance of Indexing Systems: VOLUME 1. Design, Part 1. Text Formation of Index Languages chapter Cyril Cleverdon Jack Mills Michael Keen Cranfield An investigation supported by a grant to Aslib by the National Science Foundation. Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government. - 64 - the quite specific class Interferential strioscopy to Interferometry in general, then to Photographic methods in general and then to Visualization methods in general, and so on. It will be noted that by the second reduction there is no separate class left for Schlieren photograph-. [OCRerr]nilst a distinct class is retained for Stroboscope. Yet in this subject field Schiieren photography is a decidedly more important class than Stroboscope This suggests that reduction by purely hierarchical criteria may be unsatisfr . [OCRerr]ory. When we reflect that the choice of terms within categories and the choice of the categories themselves is ultimately a matter of literary warrant, it is reasonable to assume that reduction of classes hierarchically should not be a rigid process, but should take note of the weight of literature in the different classes, so that Schlieren photography, for example, might be retained although all other sub- classes at that level were removed and incorporated in the containing class. For the single term hierarchies, this line of reasoning led to the abandonment of 'pure' hierarchical reduction and the incorporation of judgements as to the relative impor- tance of particular classes, and the noting of word frequencies in determining which classes should be retained intact at a particular level of reduction. The above example stresses the primary function of hierarchy as a recall device, whereby the index vocabulary is systematically reduced and the scope of each remaining class is consequently widened (hence the greater recall). In practice, however, by varying search programmes, hierarchical linkage allows movement in both directions - to greater precision by refining class definition or to greater recall by coarsening class definition. If specific indexing is assumed (i. e. , each document, or document-theme, is assigned to its most specific class) a search may be made in a number of different directions through the hierarchy. For example, a searcher commencing at Photorecording abf may find the amount of material there unexpec- tedly excessive and so decide to search a narrower class. This may be done, of course, independently of hierarchy, by adding a qualifier or two (e. g., moving from Photo- recording to Photorecording in'high speed wind tunnel). But it may also be done by moving down the hierarchy (e. g., to Schlieren tests). Such a decision implies that the links established by the hierarchy are permis- sive, not obligatory, and that the searcher selects from a comprehensive hierarchy just those routes he considers likely to be fruitful. It is not certain whether figures produced for a number of such searches would be useful in the sense of allowing firm generalizations to be drawn since much will depend on the subject field, and on the choice of pathways followed in searches (4) and (5) as discussed on the next page. Assume that, as shown below a, aa, etc[OCRerr]are terms in hierarchical relation: a aa aaa aab etc. ab aba abb abc etc. ae ad ae eta.