A NEW LOOK AT FEATURE CARD INDEXING

We designed our own hardware and drew up our own word list in a purely empirical way, unhampered by theory. We do not think of our word list as a language, but as a list of descriptive features, a list of only about 450 words, and this we find to be ample. It seems that there is a limit to which the feature system can subdivide a field, and beyond this limit trouble arises from cross‐fertilization. The limit, however, is more than adequate for most purposes. If greater subdivision is required it seems preferable to split the larger field into several smaller ones, and to provide a separate set of features for each of these. The general tendency for beginners is to overdo the degree of subdivision. This should be resisted in the interests of efficiency. The essence of the feature system is that it is a method for narrowing down the search; each feature that is used in retrieval divides the field by a factor which may range from about five to twenty or more, so that the convergence is very rapid. In our system we regard the author, source, form, language and country of origin as features. Proper names are also features, so that the system is self‐contained, without the necessity of separate author or title indexes.