Re: Curious coincidence

[Jorge Stolfi <stolfi@xxxxxxxxxxxxx>]

Even curiouser:

               w/o gallows    with gallows
             +--------------+--------------+
w/o tables   | 8772 (25.2%) | 9016 (25.9%) |
             +--------------+--------------+
with tables  | 8591 (24.7%) | 8423 (24.2%) |
             +--------------+--------------+
     
These are counts of tokens (word instances) in the whole majority-vote
transcription; minus key sequences, labels, unreadable/contentious
tokens, and the 326 tokens with two or more gallows.

The "gallows" are the EVA letters [ktfp], including any platforms
("ct", "cth", "ith") and isolated "e" suffixes ("te", "cthe", etc.).

The "tables" are the letters "ch", "sh", "ee", and any isolated "e"s
that are not attached to a gallows letter.

Thus, for example,
  "cthedy" and "kedy" have gallows but no tables;
  "qoedy" and "qoeedy" have tables and no gallows,
  "ykeedy" and "ykeeedy" have both gallows and tables.

By itself, the evenness of the split between the four entries may not
be too impressive. However, if you recall my three layer paradigm, the
gallows (with platforms and "e" suffixes) make up the "core" layer,
while the tables make up the "mantle" layer that always surrounds the
core when both are present.

So the table gives the presence/absence of the core and mantle layers.
I find it quite surprising that these two attributes, whose definition
was motivated by unrelated evidence, turn out to be statistically
independent and uniformly distributed, to within the range of
statistical error.

I considered counting the gallows platforms and "e" suffixes as
tables, rather than gallows modifiers. However, that choice breaks the
balance in the right column, badly -- I get 33.3% tokens with gallows
and tables, 16.7% with gallows but without tables.
(Oops, that's exactly 2:1 ! Dammit, another curious coincidence...).

If gallows platforms are counted as tables, but "e" isn't, then the
split in the right column is 27.2% with tables, 22.9% without. 
If platforms are ignored but "e" is counted as a table, the split is
31.2% with tables, 18.8 without.

Now what?

All the best,

--stolfi

PS. Gee, those VMS herbs must be *really* strong stuff. I have only
looked at digital images of xeroxes of microfilms of 500-year-old
drawings of them, and I am already allucinating...