MONO91
NIST Monograph 91: Automatic Indexing: A State-of-the-Art Report
Other Potentially Related Research
chapter
Mary Elizabeth Stevens
National Bureau of Standards
"Thus we see that the automatic elaboration of a request does, in fact, catch rele-
vant documents that were not retrieved by the original request." 1/
6.4.2 Natural Language Text Searching - Swanbon
The work in automatic indexing and related research directed by Swanson at Ramo
Wooldridge Corporation has included "indexing at the time of search" in natural language
text searching, (1960 [582, 587], 1963 [583]), the previously mentioned studies of
machine-like indexing by people (Montgomery and Swanson, 1962 [421]), and automatic
assignment indexing using pre-selected lists of clue words, (Swanson, 1963 [580]). The
last of these three major areas of investigation is the one of the greatest interest in this
present study, but the earlier experiments in machine searching of natural language texts
warrant some discussion. In his reports on this text searching project, Swanson has
specifically claimed that the methods for transforming search questions can serve as
the basis for an automatic indexing method. Thus:
"...A technique for automatic indexing can be derived immediately from a text
searching technique... it is necessary only to so organize the machine procedures
that those operations of text reduction or reorganization common to all searches
are performed only once and prior to searching in order to create directly an
automatic indexing procedure." 2/
Swanson has also claimed that if automatic searching of full text is not feasible,
then automatic indexing is not feasible, the one being prerequisite to the other. For
example:
"Clearly, if a computer technique for search and retrieval from the full text
of a collection of documents cannot be developed, then it is unthinkable that
matters could be improved by using the machine to operate on just part of
the information (a `condensed representation[OCRerr]) -- that is, on an automatically
produced index. This line of argument demonstrates persuasively that the
development of techniques for automatic full-text search and retrieval is a
prerequisite to automatic indexing. It is equally clear that a technique for
automatic indexing can be derived immediately from a text-searching tech-
nique, and thus that the two processes involve conceptually equivalent
problems." 3/
In the actual text searching experiments, a model "library" consisting of 100 short
articles in the field of nuclear physics was set up in machine-usable form. These articles
were also studied by subject specialists who rated the relevance of each paper to each of
50 questions, and assigned weighting factors representing the degree of judged relevance.
A second group of people, who knew only that the papers were in the field of nuclear
1/
2/
3/
Swanson, 1960 [582], p. 6.
Ibid, p. 240.
Swanson, 1960 [587], p. 1100.
134