Minutes of Teddington meeting

Notes on the informal VMs meeting at Denis Mardle's bungalow Teddington, Tewkesbury, Gloucestershire, UK
Sunday 21st June 1998

1. Introduction

Everyone gave a short introduction to their interest in the Vms, mostly originating from its mention in sources connected with cipher problems. Even the least active in the VMs group had studied the VMs on and off over several years.

2. Items brought by participants

Jim:

Mark:

Gabriel:

René:

Denis:

The first discussion item was a poll: Is the VMs in code or directly readable (including abbreviations or syllables)?

All agreed that no complicated code should be expected, i.e. the text should be directly readable. One dissenting voice (René) suspects a more or less significant 'garbage' component still agreeing with the rest that complicated ciphers just did not fit the statistical evidence.
It may be a language written in an invented script, and there may be word play (in the style of pig-Latin), but no more than this. A theory was advanced that the *shape* of the word matters, not the individual characters it contains.

This started a short discussion on Zipf's laws, since Gabriel pointed out that the VMs agrees with this law. Michael's view was: "many other processes also produce Zipf-like distributions". Gabriel mentioned that this was true for one known case (random strings, W. Li's paper [see ref. below], where the word lengths grow extremely large (not the case in the vms), so Li's model for Zipf's law is unlikely to apply here. Gabriel also mentioned that despite the fact that some processes may generate Zipf's distributions, there are no languages which do NOT exhibit it, and so, as it is present in the vms, we should not ignore it. (i.e. it does not prove that there is a plain text, but that there might be one).
Finally, languages have a typical way of deviating from Zipf's law at the low rank words, which is also true for the VMs.

There was general low confidence in any of the alleged decryptions, but there was much more faith in the clues given by the statistical evidence.

There was general agreement that the foliation ought to be taking into account when performing statistical tests. The order in which the pages are bound may not be the order in which they were written, causing statistically dissimilar material to be brought together. Individual bifolios are relatively homogeneous. (see René's digraph data plots, below)

3. Historical background

René gave us copies of the introduction to "Codices Vaticani Latini, Codices 11414-11709" by J. Ruysschaert, 1959, part of Bibliothecae Apostolicae Vaticanae. This specifically mentions the year 1912 in the Praefatio. It starts "Codices manu scriptos olim Collegii Romani Societatis Iesu in hoc catalogo recensitos S.Pius PP. X Bibliothecae Vaticanae anno 1912 dono dedit una cum Vaticanis Graecis 2341-2390, Vaticano Latino 13497 et Vaticano Turco 80" (Pope Pius X presented the manuscripts, which formerly belong to the Jesuite Collegio Romano, and which are described in this volume, to the Vatican Library in 1912, together with Vat.Gr. 2341-2390 and Vat.Turco 80.)

There is also the Census of Medieval and Renaissance Manuscripts in The United States and Canada by Seymour de Ricci, 1937 [ref] giving the part with the estate of Wilfred M. Voynich. This has 16 Mss of which the VMs is no. 8. And the ones bought from the Villa Mondragone are 1, 2, 8 and 13 or 14. Interestingly all the other 15 are 12th to 15th Century with two in the 12th, two in the 13th, three in the 14th and three in the 15th plus one 13th-14th, two ca 1400 and two dated 1400 and 1446. (This shows Voynich's interest). The VMs is put as 13th-15th C. Most of these Mss Voynich must have acquired early in his career. No.10 looks interesting (several hands re scientific subjects ) - first ff.1-16rare "De mysterio numerorum in Sacra Scriptura".

Finally there was a list of all known volumes bought by Voynich in the Villa Mondragone which are now in American libraries. The complete collection sold in 1912 was described as a collection of humanist works. We are left to wonder if the VMs accidentally got into this collection or was also considered a humanist work by the previous owner(s) at the Collegio Romano.

Jim had brought along his wife's copy of Cappelli. It contains many examples of shorthand which closely resemble Voynich symbols including complete words. Jim drew our attention to an example of gallows characters being using decoratively in the first and last lines of a manuscript. In this example, many different characters can be extended upwards into a gallows. It resembles very closely the gallows characters in the Voynich MS. It was remarked that in the Voynich MS, gallows are more more frequent at the start of a paragraph; this suggest that they might be purely decorative in the Voynich MS too. However, the Voynich also contains gallows in the middle of paragraphs.

Mike mentioned a connection with Wilfred Voynich and the Cambridge University Library: the library has on file correspondence with Voynich concerning the sale of some manuscripts - not the famous one, but others in his possession.

Further there was mention of 'additional material (rotograph)' in the British Library. After the meething, Jim went to the BL and confirmed: "Facs 461 is a collection of VMS photostats of ff. 1-56, dating from the Voynich/Newbold/Manly era. Facs 439 is more diverse. It has some VMS photostats in it, a collection of some VMS articles dating from 1921-22, and a collection of letters about the VMS, addressed to Robert Steele. There are letters from W. Voynich to Steele, dated 27 April 1921 and 4 June [1921?], from Newbold to Steele, 25 April, 15 August and 25 August, all 1921. Most interesting (to me) are two letters from A. W. Pollard (on British Museum letterhead) to Steele, 20 September 1921 and undated. They deplore Newbold's work. One ends "I hope we may both live to see the mystery cleared up!" Finally, there is a pencilled draft of a paper by Steele and photostat copies of typed lecture notes, I think by Newbold."

Lunch

J. Green had provided (on the net) a mapping table from Hildegard of Bingen's Ignota Lingua into a Latin/German hybrid. Several members of the group were doing further work on mapping these into modern English. It was observed that her taxonomy of the natural world is approximately that of Dioscorides (and she thinks a bat is a bird because it can fly).

There was some wondering about how she could have known about certain plants and animals (e.g. ostriches). Jim named a traveller who had written about such exotic items before her time, and whose name seems to escape us right now.

Team picture

Jim's colour slides

The colour makes it much easier to see the veins on the leaves of the plants and to separate the bleedthrough from the text. Roots tend to be drawn in red (e.g. "pharmaceutical" folios)., The jars are also in red/orange.
picture of f41v ('carrot') and overleaf page (f42r), which also appear in Toresella. These are in the two different hands.
rings with repeating cycles (f57v).
Denis offers hypothesis that the 17 chars are the consonants of an alphabet. It was pointed out that combinations like <air> and <am> occurred in these sequences, suggesting that these should be taken as single units.
Denis also pointed out that in the main sequence <k> and <t> have their own spots, but <f> and <p> share a spot and <i> and <c> share a spot
picture of plant opposite the 'mussdel' (f65v). Comments on the colours (two types of green and the blue flower)
f66r with the dead body. It became evident that f57v and f66r should be taken together:

the word <dairal> occurs in both and is otherwise rare
both pages have sequences, and Denis pointed out that it could be more than a coincidence that f57v has 17 items and f66r 34 (i.e. double).
the dead body could be related to the other four bodies on f57v

The 'wild guess' was offered that the character sequence <o,l,l,d> on f66r could represent the year 1448
In one of the "cosmological" illustrations (f68v3), most of the stars are yellow except one which is blue. It was suggested that the blue was an attempt to erase a star when the artist discovered he had drawn one too many.
Two(?) bio pages (at least: the page with the 'circus maximus': f78v, and the one with all the animals: f79v)

ladies cheeks have a red spot
'cross and ring' objects painted yellow (gold)?
water was green (f79v)
f78v there is a pool with a facade of archways: the pool above is green, but beneath the arches is blue.
water colouring applied very crudely (much like a 5-year-old with a crayon)

Zodiac image of Scorpius. Ink of 'Novembr' seems darker and tracing seems thicker (done after?) The animal itself seems in same ink as text
From one picture (which?) it became clear that when writing <sh>, the curl was written first
It was observed that in places the Voynich text fills the available space too neatly. e.g. in the folio where there are two words above each nymph (f75v), the lengths of the words are highly correlated: a big gap between nymphs gets two long words, and a short gap gets two short words. There are many other examples. This suggests that the text is unrelated to the pictures, and was just written around the pictures wherever there was space.

4. Statistics

Jim and Denis outlined hidden Markov modelling. Hidden Markov models treat a symbol sequence as noisy observations of a Markov chain whose state is hidden. There is an iterative algorithm for deriving a state-transition matrix and a noise matrix from a symbol sequence. You choose a number of states, and the algorithm finds a model with that number of states. If the source really is Markov, the goodness of fit of the derived model reaches a plateau as the postulated number of states reaches the true number of states. The algorithm can get caught in local optima, so it is necessary to run it several times with different starting points. The hidden states can be relabelled without affecting the observed result, so with k postulated states there are k factorial copies of each local optimum.

Denis presented results of hidden Markov models applied to Linear B and other languages. Two-state approximation show the space/letter distinction. Three state models for alphabetic languages (not Linear B!) show the space/vowel/consonant distinction. Larger number of states show word-stem/suffix distinctions.

Jim has done some work on applying hidden Markov models to the Voynich MS in 1991. His initial results broadly confirm those obtained by Jacques Guy using Sukhotin's algorithm (if this distinction is highly probable it de-emphasises the consonant only theory - DM).
It was felt that this approach would let us put the stem-suffix analysis of the Voynich MS on a more systematic footing.

Some numericalresults shown:

To be added to this text are the occurrence of <ed>, the fact that the foldouts are in the middle and the possibly special place taken by the inner bifolio of the bio section.
Michael argued that the plots show two ellipsoids in N-space (one for each author) and René has projected them from a viewing angle which makes the ends of the two ellipsoids overlap." René added that this is possible, but that the N-space clouds are not separated in any dimension.

5. Summary

Not all the Agenda could be completed, as one would expect, but the meeting was clearly useful and the time (11:00 - 17:00 approx) with a break in the sun for lunch seemed to go only too quickly.

References:

J. E. Fletcher: Johann Marcus Marci writes to Athanasius Kircher. Janus, Leyden, LIX (1972), pp. 97-118

J. Godwin: Athanasius Kircher: A Renaissance Man and the Quest for Lost Knowledge. London, Thames and Hudson, 1979.

W. Li. Random texts exhibit Zipf's-law-like word frequency distribution. IEEE Transactions on Information Theory 1992; 38(6): 1842-1845.

Seymour De Ricci, with the assistance of W. J. Wilson: Census of medieval and Renaissance Manuscripts in the United States and Canada, 1937. Kraus Reprint Corporation, New York, 1961

Jose Ruysschaert: Codices Vaticani Latini, Codices 11414-11709, Schedis Henrici Carusi Adhibitis, Recensuit Jose Ruysschaert, Bibliothecae Vaticanae Scriptor, MCMLIX (1959)

Frank Smyth. (article title?) In: The Unexplained: mysteries of mind space & time. (weekly publications) Orbis, London, pages 1381-5 and 1418-20, 1980-. (paper with a portrait gallery and the unfortunately bad-quality nine-disk picture)

VMs meeting in UK, 21 June 1998