Journal of Information Science, Vol:19 (2) p.164-165, 1993
Mechanical indexing, structural linguistics and information retrieval

Dear Sir,

I am most grateful to you and the Journal of Information Science for publishing my article on "The relationship between mechanical indexing, structural linguistics, and information retrieval," (Vol. 18, No. 5, 1992). As you say in your introduction, the paper was written over 33 years ago. You ask why it was not offered for publication at that time. Quite simply, other academic and personal priorities intervened. By the time Naomi Sager began to publish her work, I had not only forgotten about my manuscript, but was out of touch with Professor Harris and the Structural Linguistics Department at the University of Pennsylvania. Not only had I gone on to complete my doctoral dissertation [1], but I was involved in research on the Genetics Citation Index from 1961-1964, followed by the introduction of the Science Citation Index ® in 1964.

You go on to say that a second referee claimed that "all of the ideas put forward by Garfield have been researched, elaborated and published by Naomi Sager, over ten years ago". This is an uninformed, offhand comment with which even Professor Sager disagrees, as is verified in the enclosed letter she wrote [2] after reading my article in the Journal of Information Science.

I suggest you publish her entire letter, but the relevant quote is that "it is quite fair to say I [Sager] have not published anything directly related to the indexing or cataloging process However, as an editor for the Journal of Information Science, I asked her to cite those papers she considers most relevant.

As an extension of my comments on the linguistics-indexing connection, it is relevant to mention the recent keynote address of Herbert A. Simon at the annual meeting of the American Society for Information Science in Pittsburgh[3], in which he discussed his work in artificial intelligence. In particular he referred to the paper on auto-abstracting by H.P. Luhn [4] which I also cited. A proper postscript to my paper and/or Luhn's would be that we were working in artificial intelligence but didn't know - a rose by any other name is not necessarily just as sweet!

All of the above demonstrates differences in research cultures. Cognitive science researchers, like Herb Simon of Marvin Minsky, described their field as "Artificial Intelligence". Those of us who came from the world of "indexing," (also known as documentation, library science or information science) called the field "automatic indexing" or "auto-abstracting". Since the latter has been outside the mainstream of computer science (unlike the work of G. Salton) it has received less recognition. Computer algorithms for playing chess intrinsically have greater popular appeal. Nevertheless, while computers can now beat chess masters, there still are no computer algorithms that can fully match human indexers as described by the examples in my paper and by the pioneering work of John O'Connor [5]. That applies not only with respect to a posteriori indexing by knowledgeable users, but also with respect to a priori[6] indexing procedures as performed, e.g. by medical indexers. Computer algorithms can produce indexing equivalent to, or sometimes exceed, but are not identical to human indexing.

That brings me full circle to the two main points of my paper which were to show not only the potential power of linguistic analysis but also its limitations, including especially the metatext issue. I met Harris at the time my first paper on citation indexing was in press [7]. When I de scribed citation indexing to him, he shrugged it off as a problem of metatextual analysis. I asked him how one could index any scientific paper algorithmically if the full meaning of its content was dependent upon the other papers and books it cites? From these early considerations and later experience with publishing citation indexes, I was convinced that we had to combine natural language and linguistic analysis with citation indexing. The first expression of this was Permuterm indexing [8]. More recently, this has culminated in Keywords Plus TM[9] in which algorithmic indexing is based upon the use of titles in cited and citing papers.

Yours faithfully

Eugene Garfield
Founder & Chairman Institute for Scientific Information
Philadelphia, PA USA


[1] back to text E. Garfield, Chemico-linguistics: computer translation of chemical nomenclature, Nature, 192 (1961) 192.
[2] back to text N. Sager, personal communication, December 1992.
[3] back to text H.A. Simon, Keynote address to ASIS, October 26,1992, Pittsburgh, PA.
[4] back to text H.P. Luhn, The automatic creation of literature abstracts, IBM Journal of Research & Development 2 (1958) 159-165.
[5] back to text J. O'Connor, Mechanized indexing methods and their testing, Journal of the Association of Computing Machinery, 11(1964) 437-449.
[6] back to text SM. Humphrey, Indexing biomedical documents: from thesaural to knowledge-based retrieval systems, Artificial Intelligence in Medicine 4 (1992) 343-371.
[7] back to text E. Garfield, Citation Indexes for Science, Science 123 (1955) 108-111.
[8] back to text E. Garfield, The Permuterm Subject Index: an autobiographical review, Journal of the American Society for Information Science 27 (1976) 288-291. pdf file available
[9] back to text E. Garfield, Keywords Plus ISIs breakthrough retrieval method, Essays of an Information Scientist 13 (1990) 295-304.  pdf file available