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. -78 - different contexts in which a term appears are gathered together as qualifiers of that term. Although in a real-life situation the first method provides easier access to these further relations, this advantage was not significant in the test environment. The test collection subset was relatively sma[OCRerr] and a fully rotated A/Z index was easily producable by clerical labour. In any case, such an index was necessary for other reasons, too, as will be described. Moreover, although theworking out of full al- ternative hierarchies would have involved a considerable effort, there was no guarantee that more than a small fraction of them would ever be used, since only those hierarchies relevant to the terms occurring in questions would be required. The assumption above is that such an A/Z index will automatically disclose the existence of other possible hierarchies. Indeed, it is difficult to see how such ad- ditional hierarchies could be economically developed unless we are guided by the literary warrant afforded by the actual occurrence of the terms concerned in con- junction with these other contexts, in which case the A/Z index automatically picks them up. Nevertheless, further connections were indicated by references in the schedules wherever it seemed desirable, particularly where it seemed that the A/Z rotation of terms might still fail to show the connection; e.g. Small disturbance theory, subordinated to Disturbance, contains a reference to Boundary layer theory to which it is also relevant. Streamlines, subordinated to Flow elements, has a reference to Relative stream surfaces (in Compressor "flow phenomena} to which it is also generic. Or, within a given class, references were added to link concepts occurring in dif- ferent arrays; e.g., Performance discontinuities in the Performance facet of Com- pressors contains a reference to Stall and Surge in the Flow facet of the same class. It may be noted that in a real-life classified index, the A/Z index usually shows even those connections just exemplified, since its entries contain more qualifying material (providing further information regarding the context} than the test index, where the concepts gave the sole element of precoordination. For example, a docu- ment dealing with Small disturbance theory in the context of Boundary layer theory would produce rotated A/Z index entries: Small disturbance theory: Boundary layer theory Boundary layer theory: Small disturbance theory and these establish the connection which in the test collection had to be established by references. The significance of multiple hierarchical linkage as an element in the recall performance of generic hierarchical linkage generally is probably not very great. Most questions impose a particular context of their own and the likelihood of relevant material being found in radically different contexts of the particular terms of the question is probably small. For example, a question on the kink in the surge-line of a multi-stage axial compressor imposes a context on the notion of 'surge'; clearly, documents indexed under Surge as a general concept should be examined, but it is unlikely that extended examination of the classes flanking Surge in the general hierarchy of Aerodynamic processes would be very fruitful. The A/Z Index (see Appendix 5.5) In order to provide for multiple generic hierarchical linkage as discussed above, and for other reasons, a rotated A/Z index of the concepts was now produced; e.g.,