Journal of Documentation 31(3):216 September 1975
Letters to the Editor
WHAT IS AUTOMATIC INDEXING?
Dear Sir,
In a recent review on ‘Automatic Indexing’ in Journal of Documentation 30, 393-432 (1974), Sparck-Jones provides an interesting view of her conception of automatic indexing and classification. It is interesting because it is typical of most work in the so-called and improperly named field of ‘automatic classification.’ Most people have finally stopped equating algorithmic with computerized. Why not take the same step for automatic indexing and call it algorithmic indexing? This is really what we are all talking about.
While the computer has played a prominent role in facilitating production of the Science Citation Index, it is by definition an algorithmically compiled index. Why then do Sparck-Jones and others persist in ignoring the reality of the SCI as the largest extant automatically, that is algorithmically, indexed collection available. If this assertion is accepted, then the statement that "there is little hard evidence as to the value of citations in an automated system, particularly as substitutes for other modes of indexing, as opposed to additional keys" becomes absurd. Sparck-Jones seems never to have heard of the Automatic Subject Citation Alert 1. The ASCA system has been operating successfully for over ten years. Simkins indicated years ago2 the ‘evidence’ for ASCA value. This was recently confirmed by Parry et al.3 Is not the successful unsubsidized operation of a real-life system not considered ‘hard evidence’ of value?
In the United Kingdom, OSTI supported tests of ASCA and found considerable ‘value’ in the method. Sparck-Jones seems oblivious of these data.
If the SCI is not an automated or algorithmic system, then neither is any one of the other systems she describes. All require a set of rules for keying or selecting data. What evidence will Sparck-Jones accept as confirmation of the value of citations? Without evidence, she claims that Salton’s tests were not conclusive. On the other hand, she does not cite his own positive conclusions about the use of citations: ‘ . . . in some circumstances, citations are more effective for retrieval purposes than other more conventional terms and concepts.'4
As I stated in a recent paper at the Third International Conference on Classification Research in Bombay, there is a reluctance in the ‘classification’ establishment to accept citation indexing as automatic indexing because it is all too ‘simple’. It is not rationalized on linguistic, episternological, or other esoteric grounds. Is it therefore less valid?
The difficulty with the classification research community is that it stubbornly refuses to recognize the real-world value of SCI or other systems. They are used every day by thousands of clients who do not require philosophical analysis to measure value received. What theorists should be rigorously seeking is why it does work and what its fundamental implications are for linguistic and other studies.
E-mail: garfield@codex.cis.upenn.edu
REFERENCES 1. back to text GARFIELD, T. and SHER, I. H. ISI's Experiences with ASCA—A Selective Dissemination System, Journal of Chemical Documentation, 7, 1967, pp. 147-53.2. back to text SIMKINS, M. A. Retrieval of Information from the Recent Literature of Medicinal Chemistry, Chemistry and Industry 5, 3 February 1968, pp. 146-50.
3. back to text PARRY, A. A., LINFORD, R. G., and RICH, J. I. Computer Literature Searches—A Comparison of the Performance of Two Commercial Systems in an Interdisciplinary Subject, Information Scientist, 8, 1974, pp. 179-87.
4. back to text SALTON, G. Automatic Indexing Using Bibliographic Citations, Journal of Documentation 27, 1971, pp. 98-110.