CRANV1P1 General Considerations 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. Means of control Field of use 6. Generic descriptors. 7. Specific and generic descriptors linked hierarchically. 8. Multiple generic links for each specific descriptor. 9. Categories of descriptor, form- ing facets. 10. Semantic factors to represent subject terms. 11. Correlation of descriptors. 12. Weighted descriptors. 13. Interlocking sets of descriptors. 14. Regulated sequence of descriptors. 15. Interfixing descriptors. 16. Role indicators. 17. Relational terms. Many mechanized systems. Classifications, thesauri, some subject head- ing lists, some mechanized systems. Some classifications, subject heading lists, and thesauri, a few mechanized systems. Faceted classifications, some mechanized systems. To some extent in faceted classification, the W.R.U. system, mechanized patent office systems. Many alphabetical indexes, some classified catalogues, all mechanized systems. Some experimental computer systems. Alphabetical indexes, classified catalogues, computer systems. Alphabetical indexes, faceted classifications, fixed-field punched cards, some computer systems. Mechanized patent office systems. Some faceted classifications, some mechanized systems. Alphabetical indexes, some faceted classifi- cations, some mechanized systems. All the results of Cranfield I pointed to only one conclusion. Whereas one could evaluate the performance of an operational information retrieval system and find how the index language being used affected the performance of the particular system under investigation, it was not possible to do any basic research on index languages by this method, for there are so many uncontrollable variables in any operational system that comparison of index languages is impossible. It has to be admitted that this view is not generally held, since one finds a new investigation which has the objective of comparing various UDC operational systems with other operational systems using different types of index languages. In that this results in even more variables than existed in Cranfield I, it is difficult to see how any valid data concerning the UDC can be obtained. On this point Richmond is in complete agreement, for she writes (ref. 8) "System evaluation by comparison testing is essentially a negative operation", and again, "Comparison with other systems does not answer problems arising from the weaknesses of this system. In each case, the faults are internal and only obliquely subject to evaluation by comparison with other systems". To make advances in knowledge regarding index languages, what was now re- quired was a laboratory-type situation, where, freed from the contamination of operational variables, the performance of index languages could be studied in isolation. While such an approach waa unusual in 1961, at least two other organizations have also established similar conditions, namely the Centre for Documentation at Western Reserve University and the Computation Laboratory of Harvard University. The methods used at Cranfield to establish this situation are considered in the following chapters of this volume.