I once read a paper contending that the "Thomas More" codex was
done by W.Shakespear, because the usage of dozen pairs of "skip
words" (eg. a/an, all/any, better/best,...) gives no significant
differences vs all the alledged authentic works of Shakespear.
I wondered on reading :
1. why not try other works ?
I tried the same test with "Othello" vs "all the tradgedies of S",
and "King John" vs "all the historical plays of S", and got
significant results. The author should have argued that Othello
or King John is not authentic.
2. why not try other authors ?
I tried the same test with some works of Conan Doyle's Sherlock
Holmes series, eg. "The hound of the Baskervilles" vs "The study
of scarlet", etc. Some of them gave significant results.
I agree that tests may vary according to the text examined, so this
may not be a serious problem.
3. why "non-significant" means authentic ?
I tried the test with Doyle vs Shakespear. Totally, they are quite
significant, but some pairs give still non-significant results.
This means that, some "non-significant" indexes are REALLY non-
significant :-), ie. they are constant across authors and centuries,
and have nothing to do with the authorship.
The author of the paper should have looked for indexes "constant in
one author and various among authors", instead s/he has found something
just constant.
The authorship test are based on stylistic indeces, which can also be used
for dissemination of different works done by an author.
The hyphothsis "the difference in one author is narrower than among authors",
may sound too optimistic.
# But, I am still quite interested in the statistic methods for the
# authorship, and would like to hear much more about it.
Masayuki TOYOSHIMA / Dept. of Japanese linguistics, Faculty of Letters,
Hokkaido university, Sapporo 060 Japan, FAX +81-11-726-0919 (24h)