In 1936 a concordance to James Joyce's Ulysses was created on
cards and circulated by mimeograph. Harvard's George Kingsley
Zipf noticed an interesting thing about the statistical tables
generated from this process: an exact inverse correlation between
frequency and appearance of words existed. Thus, the commonest
word ("the") occurred 16,000 times and--as if to compensate--there
are exactly 16,000 words that appear only once! And so on. A
log-to-log graph of frequency versus occurrence is a straight
slanting line.
-- [noted by Hugh Kenner, BYTE, Oct. 1989, p.362]