PRELIMINARY STUDIES FOR MACHINE GENERATED INDEX VOCABULARIES.

This article presents a preliminary studies for machine generated index vocabularies. The vocabulary of descriptors was generated by first finding all possible pairs of words whose rate of co-occurrence within the same document was abnormally high. Then by various alternative rules, these "significant word-pairs" were amalgamated into groups of 3 to 8 words. Suitability was measured by retrieval trials, or possibly some function representing information. Every lengthy procedure in the formation of word-groups ought to be suitable for computer programming. Probably also, any particular word should be in the group containing the most words with which it forms significant pairs.