RE: More Chinese stuff: Shennong Bencao

[Jorge Stolfi <stolfi@xxxxxxxxxxxxx>]

As I observed in a previous message, there are several intriguing
resemblances between the Sennong Bencao Jing (the classical Chinese
pharmacopoeia), and the VMS "recipes" section (f103r--f116r). The
number of recipes seems to be compatible (365 in the former, 323--380
in the latter); the average number of tokens per recipe -- ideographs
or "words", respectively --- is fairly similar (36 versus 33); the
histograms of recipe lengths have many similarities; and both texts
have a fair number of word doublets (41 and 80-90, respectively).

The difference in the average token counts could be easily explained 
by accidental joining and/or abbreviation of tokens, which 
almost certainly occurs in the VMS.  The increased number of doublets
may be due to homophony -- two distinct chinese characters having the 
same sound, at least to the author's ears.

A more worrisome difference is that the VMS recipes section uses
over 2800 distinct words, while the Bencao uses only 1100 distinct
ideographs.

The obvious next step was to check whether the sequence of recipe
lengths (token counts) is the same in both manuscripts, except for a
36/33 scaling factor. That would be too easy, of course. No luck: as
far as I can tell, the two sequences are quite unrelated: 

  http://www.ic.unicamp.br/~stolfi/voynich/misc/bencao-tk-counts.png
  http://www.ic.unicamp.br/~stolfi/voynich/misc/vstars-tk-counts.png
  
Th "zero length" recipes in the above images indicate section boundaries in the 
Bencao, folio boundaries in the VMS.

I cannot see any resemblance even if the sequences are smoothed out:

  http://www.ic.unicamp.br/~stolfi/voynich/misc/bencao-tk-counts-diff11.png
  http://www.ic.unicamp.br/~stolfi/voynich/misc/vstars-tk-counts-diff11.png

Here the bottom trace is a 9-element, binomial-weighted running average
(i.e. the sequence convolved with a 9-element binomial distribution).
The top trace is a 5-element average divided by that 9-element
average.

An obvous explanation for those discrepancies is 
that the two texts are unrelated, and the resemblances noted above
are meaningless coincidences. 

However... 

For 50 years at least, people have been guessing (based on the
contents of the other sections, and the organization of contemporary
herbals like Culpeper's) that f103-f116 contains a list of medical
recipes. Yet, to my knowledge, no one has managed to find any Western
recipe list that shows even a vague statistical resemblance to that
section. Surely it wasn't for want of trying. Therefore it is hard to
dismiss the resemblances to the Bencao as "mere coincidence".

Moreover, *if* the manuscript is indeed a phonetic trannscription 
of an East Asian medical text, as claimed by the Chinese Theory,
then the most natural candidate for the source of the recipes
section would be precisely the Shennong Bencao.  By 1500, that 
book was still the "bible" of Chinese medicine, and even today it 
seems to be the most quoted source in herbal medicine pages,
Western ones included. 

Thus, it seems worth spending some more time investigating
this possible connection, even if it calls for some 
questionable "excuses". 

The difference in vocabulary size, for instance, could be explained by
assuming that about 15% of the VMS tokens (about 1 in 7) are affected
by some form of spelling variation. This estimeate does not seem tooo
high. In European texts of the eriod, one often finds the same word
spelled in two or more ways on the same page, often in the same
sentence --- on *printed* books in the author's native language.

The lack of correspondence between the sequences of recipe size could
be due to rearrangement of the recipes. Alternatively, the VMS recipes
may have been dictated from a translation or adaptation of the Bencao
to some other monosyllabic language, such as Vietnamese or Tibetan,
with a different grammatical structure. Such a translation would have
been writen in Chinese characters, but the number and order of the
words would vary.

Yet another possibility is that my reference text for the Bencao
(which was prepared by prof. Kenji Kobayashi of Japan) is not the
Chinese original, but an ancient translation into Japanese --- which
at the time was written entirely in Kanji.

Indeed, the Bencao recipe size plot above shows a clear shrinking
trend along the manuscript, from about about 40 chars/recipe at the
beginning to 30 or so at the end. It seems that the scribe's style
gradually became more succint as he progressed on the manuscript. This
feature is not expected in a book that presumably was composed bit by
bit over many years (or in a simple copy thereof.)

Thus I would say that the jury is STILL out on this question...

All the best, for now...

--stolfi