A. R. Meetham’s research while affiliated with National Physical Laboratory and other places

A semantic group of words is one whose words are associated in meaning, in some way definable by the lexicographer. A thesaurus is a large collection of words, each of which is assigned to one or more semantic groups. The importance of thesauri for applications in library information storage and retrieval is evident from the number of centres interested in them. (No less than 24 centres are listed in the index of Current Research in Scientific Documentation No. 13, and thesuri in 10 disciplines are mentioned specifically.) The present paper describes aspects of an attempt to generate groups of words by purely mechanical means, which seem to have strong semantic connections within groups and which, therefore, are of potential value as thesauri. The lexicographer plays no part at all, and therefore no semantic processes can be used. However, the machine can make very many more binary decisions than a lexicographer, and there is the possibility that a machine generated “thesaurus” is at least as valuable for document processing as a man-made one.

The report is a continuation of “Preliminary Studies for Machine Generated Index Vocabularies”, A. R. Meetham (1963) Language and Speech, 6, 22. It assumes that documents employ words from particular groups connected with their subject matter, and discusses four methods, two of them new, for finding the groups. They are picked out from a word list by using a word-word binary matrix to represent the associations between pairs of words. In an evaluation, the method which consumes least computer time turns out also to be the best.

The studies form part of a programme in which a computer will be used to generate an “index vocabulary” appropriate for the indexing and retrieval of documents in various scientific disciplines; the computer will then index a suitable collection of documents and retrieve documents in response to specific enquiries. The present paper describes how such a system works for a particularly simple text. Then, ten evaluation methods are put forward, and used to assess individual descriptors, indexing rules, and the system as a whole. In this exercise the evaluation methods themselves are on trial and four are selected as being more generally useful than the others. For evaluating whole retrieval systems a type of random enquiry is used for which the perfect answer is computable; relevance and recall ratios can then be objectively determined.