MONO91
NIST Monograph 91: Automatic Indexing: A State-of-the-Art Report
Problems of Evaluation
chapter
Mary Elizabeth Stevens
National Bureau of Standards
"For the documents in our system, we estimate that processing time will be
about 20 seconds per thousand words .. . The cost is approximately $3.50
per minute when averaged between prime and extra shift." 1/
This means that the cost of processing a 3, 000-word document would be $3.50 , exclusive
of the costs of keypunching the input text which, conservatively estimated, costs not less
than 1-2 cents per word. 2/ Swanson similarly assumes either that machine-usable text
is already available or that editing and keystroking efforts are separate costs in arriving
at an estimate of $1.00 per item for automatic indexing.[OCRerr]3/
These quantitative estimates bear out the more subjective conclusions of such
investigators as Bar-Hillel, O'Connor, and others. Examples are:
`11t is very likely that manual uniterm indexing by cheap clerical labor will still,
on the average, be qualitatively superior to any kind of automatic indexing, and
it is very unlikely that the cost of automatic indexing will ever be less than this
kind of manual uniterm indexing, unless the automatic indexing is to be of such
low quality as to totally defeat its purpose." 4/
"Most of these techniques require that the full texts of documents be in machine
readable form. At present this usually requires keypunching which is much
more expensive than a specialist's indexing efforts."
1/
2/
Gallagher and Toomey, 1963 [205], p. 52.
`[OCRerr]ompare, for example, Ray, 1961 [496], p.55; Swanson, 1962 [584], p. 470:
The cost is roughly one or two cents per word which by standards of what is
normally spent for even the most thorough indexing and cataloging, is
exorbitant." Mersel and Smith report 1964 [415], p. bA) typical TRW costs
of keypunching as two cents per word for Russian technical text, and one cent
per word for English. They also cite cost figures as low as half a cent per
word at the CIA-Georgetown Keypunching Center in Frankfurt and at IBM, but
this is exclusive of overhead and computer processing (e.g., editing program)
costs, so that the one cent figure appears minimal as of today. However,
Kochen reports (1963 [327], p. 7): "While keypunching of text cost roughly one
cent/word, new means for recording spoken (and written) text using a steno-
keyboard tied to a photodisc storing a Stenocode -English dictionary could possibly
reduce the cost to 1/3-cent per word."
3/
Swanson, 1962 [584], p. 471.
4/
Bar-Hillel, 1962 [35] , p. 418.
O'Connor, 1963 [443j', p. 1.
155
5'