While going through old voynich.net posts (courtesy of Rich SantaColoma, thanks!) in my hunt for previous insights into how the zodiac nymphs were ordered, I found an unexpected ally: on the old mailing list in Jan 2003, a now all-too-familiar face asserted that the nymphs were ordered inner ring first and clockwise, starting from 10 o’clock. Though by February 2003, that same person then got tangled up in Rene Zandbergen’s George of Trebizond theory, and lost focus, which was a shame.

That person was, of course, *checks notes* a certain Mr. Nick Pelling. *sigh*

Why did I think inner ring first? I can’t reconstruct my argument from back then, but I think it was something to do with ink strength on some pages (when going from inner to outer). But even so, I’m now happy to instead rely on Stefano Guidoni’s far stronger argument: Stefano pointed out that the barrels finish in the inner ring of Taurus II, which suggests that the barrels were abandoned halfway through.

As far as clockwise or anticlockwise, I’m reasonably comfortable with the fading ink on the outer nymph ring of Taurus I as evidence for clockwise:

So, if what we’re looking at is actually a saints calendar, and we can now number the days, and we can make a tolerably good guess at which month each one represents (and because of the 29 stars in Pisces, I’m much more minded to read that as February, rather than the March that has been inelegantly added), then we can try to work back to the month and a day linked to each zodiac nymph.

But… most of the nymphs look basically identical (or perhaps rather our modern understanding of saints is not subtle enough to help us tell them apart). So I thought it would be good to look at some nymphs who visually stand out from the crowd, and see if there might be anything interesting about their reconstructed date…

Beardy man, 22nd February

This appears on the Pisces page in the outer ring, on the bottom right. I believe it’s the only bearded zodiac nymph, but please feel free to tell me if I’ve missed any others.

22nd February is the Feast of Cathedra Petri (the Seat or Throne of St Peter), one of the oldest feasts in the Catholic Church. Because St Peter was considered to be the first Bishop of Rome, his seat (by synecdoche) “signifies the episcopal office of the Pope”. Inevitably, there’s a Wikipedia page on it.

Is this beardy guy meant to somehow be St Peter? Maybe, maybe not: but that’s as far as I’ve got.

Stray barrel nymph, 7th August

On the Virgo-roundel zodiac page, there’s a single nymph standing in a stray half barrel:

This looks vaguely to me like a medieval half-barrel bath tub:

Incidentally, the last image is Melusine having a wash, from a JSTOR daily post. But do I have the faintest idea why there’s a half-barrel bathtub associated with this day? I most certainly do not, sorry.

As to the date (7th August): what I quickly found is that many saints are associated with each day (that is, there seems to be a huge oversupply of saints relative to days of the year).

However, if you think that Milan might be connected to the Voynich Manuscript in some way, then you might be interested to know that 7th August was the day linked with St Faustus, a Roman soldier martyred in Milan in ~190. To be fair, there’s almost nothing known about him historically, so the Church quietly dropped him from the Roman Martyrology list in 2001.

Grassy nymph, 26th May

The unusual thing about the nymph at the top left of the Gemini page is that she appears barefoot, and apparently standing on some grass. So… what might that all be about?

What I quickly found out was that 26th May is the day of Madonna of Caravaggio. The story (there’s a better page in Italian here) is that on 5pm on 26th May 1432, an abused peasant wife Giannetta de’ Vacchi was collecting grass for her animals in Mazzolengo meadow near Caravaggio (near Cremona in Bergamo, 20 miles east of Milan), when the Madonna suddenly appear to her in a bright light. Mary told her that She was displeased about her husband’s drinking (and that She would stop that): and also that Giannetta should convince Milan and Venice to halt their war. The Madonna then touched the ground with Her foot, and a spring appeared.

Despite initial disbelief from the people, the spring (and Mary’s footprint) quickly started to attract visitors and pilgrims. Giannetta ended up meeting both Filippo Maria Visconti (the Duke of Milan) and Francesco Foscari (the Doge of the Serenissima); and in 1433 the war stopped. Duke Filippo Maria Visconti wanted a church built near the spring (this was consecrated on 20th December 1451, supported by the Sforzas). And in 1475 this modest church was replaced by a Sanctuary, which still stands. Even today, you can go down into an undercroft beneath the Sanctuary and draw water from that spring.

Might the grassy nymph be signifying the Madonna of Caravaggio? Nothing’s for sure (yet), but I think it’s a strong possibility, particularly if the Voynich Manuscript was made near Milan from 1432 onwards.

I’ve spent some time recently revisiting the Voynich Manuscript’s labelese, as well as its Pisces zodiac roundel page, and thinking about how that might relate to February. However, making all of these parallel strands “land” at the same time has proved difficult: even if the zodiac labels are some kind of cisiojanus “syllable soup”, we still have many practical problems linking everything together into one solid decryption.

Still, having now spent some time putting February’s “Bri pur bla sus” saint’s days and festivals under the microscope, it’s becoming apparent that many of these Christian saints were martyred virgins: and so perhaps the whole notion of oddly-angelic naked nymphs isn’t as far away from the subject matter as you might at first think.

Moreover, having thought about the really important feast days associated with men, I’m coming round to the idea that perhaps these may be connected to the few “male nymphs”. I’m thinking specifically about whether the beardy breastless nymph below might be connected with February 22nd, Cathedra Petri [the Feast of St Peter’s Seat].

So I’m now coming round to wonder: if the (relatively few) male nymphs in the zodiac section are broadly linked to specifically male feast days, might we be able to use them to reconstruct the nymph numbering? (i.e. which nymph is linked with which day.)

But before launching into that, I thought it would be good to see what people had previously posted on this general topic.

Notes on Nymph Numbering

D’Imperio mentions (3.3.3) that Peterson noted that some of the nymphs might be male: but doesn’t seem to mention trying to reconstruct the correct order of the nymphs.

Going through the voynich.net archives reveals various observations:

  • Rene Z [15 Aug 1997]: “The nymphs were drawn from the inside ring outwards, with the text added either immediately or afterwards. I think there are two possibilities for the order: either starting near 00:00 and going clockwise or starting near 09:00 and going against the clock, this from observing where the nymphs are more cramped together (especially the inner circle of Sagittarius).”
  • Rene Z [15 Aug 1997] “There is one nymph in Gemini without a label. I would favour the idea that this was a simple oversight. There is also one nymph without a star somewhere…”
  • John Grove [05 Oct 1997]: “In the June and December pages, the first nymph outside the circle has a ‘carpet’ under her feet. If you read the calendar from the inside out (as I have had a tendancy to do), these two nymphs occur 5 (for June) and 4 (for December) — days? — before the end of the zodiac month.”
  • Rene Z [1 Oct 1998]: “In the zodiac section, the standing nymphs all have their right hands either pointing backwards or placed on their hips.”
  • Rene Z [18 Jan 1999]: “About 1 out of 6 of the standing nymphs have their hand pointing behind them, not on the hip. But for nine-pointed stars this fraction is zero. I checked that the probability of this is 0.02 if this was just due to chance.”

One thread also suggested looking here at a woodcut in: Paul Heitz (Hg.), Einblattdrucke des 15. Jh., Bd. 18: Richard Schmidbauer, Einzel-Formschnitte des 15. Jh. in der Staats-, Kreis- und Stadtbibliothek Augsburg, Straßburg 1909, Taf. 9. Wilhelm Ludwig Schreiber, Handbuch der Holz- und Metallschnitte des XV. Jh., Leipzig 1925-1930, Nr. 1883a. However, Google Books didn’t seem to have a copy of this, alas.

I also found a VoynichViews blog post that highlighted a specific Sagittarius nymph holding her star downwards.

Please feel free to point me towards any posts that specifically discuss issues around determining / reconstructing the correct order of the Voynich zodiac nymphs, because (for example) I had no luck finding anything on voynich.ninja.

However, most of the interesting thoughts on the (old) list were in a single 1999 post by Jorge Stolfi: he wondered if he could discern any visual clues that signalled which was the first nymph of each zodiac page. So I decided to copy the whole post here as a section in its own right…

Jorge Stolfi’s thoughts [21 Jan 1999]

(Here’s what Jorge Stolfi posted on the subject of nymph numbering back in 1999):

Yesterday I thought that those nymphs might mark the starting point
for reading each ring of stars. Now that I have looked at those cases
with some care I am not so sure. Anyway, here are the cases that I
could see:

70v2 Pisces

 There is one nymph with both arms raised at 00:15 in
 the inner band.

 I would say that this is the most likely starting point for
 the inner star sequence, which runs clockwise (agreeing with
 the text). Thus I think that the inner parade begins with
 Miss Otalar (stretched arm), and ends with Miss Otaral
 (facing clockwise)

 This is another anomaly of Pisces, since in the
 other diagrams the starting point seems to be around 10:30. I
 suppose tha the last nymph at 11:30 was reversed so that it
 would face the "honor spot" at noon.

 The starting point of the outer band is not so obvious. I
 would say it is near the top, too, but it could be before,
 after, or in the middle of the four "baby" nymphs.

f70v1 Aries “dark”

 Here all nymphs have the right hand on the hip; several have
 the left hand down too. My guess for the starting point is at
 10:30 in both bands, i.e. Miss Otalchy (the Tar Am Dy) and
 Miss Okoly. Note that they (and only they) are holding their
 high enough to intrude into the surrounding text ring.
 Moreover Miss Okoly is wearing a striped sleeve (or
 whatever).

 Note again that the label at 06:00 is not obviously
 associated with any star, so it must be attached to one of
 the nymphs. I would say that, going clockwise, each label is
 associated with the preceding nymph.

f71r Aries light

 Here all nymphs have the same pose: right arm
 on the hip, left arm up and holding the star.
 The stars have no tails, except for the outer 04:30
 one that has a very short one.

 The starting point for each text ring is clearly marked by
 the "notched square" device, which occurs in other cosmo
 diagrams, presumably with the same function.

 As I argued in my previous message, this is the zodiac page
 with the most "primitive" style.

f71v Taurus “light”

 Here too all nymphs have the same pose. I see no obvious
 "start" marker for the nymphs, except perhaps for the
 decorated dustbin of Miss Otalody, the inner nymph at 00:00.

 However the outer text ring has a wider gap at 10:30 (the
 "standard" starting place), with a centered dot which may be
 the last vestige of the notched square symbol.

 To my eyes, the style of this page is only a bit less primitive than that
 of f71r.

f72r1 Taurus “dark”

 The outer nymph at 02:30 has her right arm stretched back and
 down; all the others have the right hand on the hip or inside
 the dustbins.

 There are no obviosu start markers that I can see, but the
 reproduction I have is unreadable around 03:00. There is
 anextra wide gap in the inner parade around 10:30, but that
 may be a consequence of the "cigarette hole" and its visual
 pun. Other plausible candidates are the nymphs at 00:00, Miss
 Otchoshy and Miss Oaiin Ar-Ary.

 I would say that the figures on the outer band of this
 diagram are the first attempts by the artist at drawing
 full-body naked women.

f72r2 Gemini

 My copy is almost illegible. I can see on the outer band one
 naked nymph at 10:30, Miss Okar-Aldy, with the right arm
 stretched out. That seems to be the "standard" starting
 position in several other diagrams.

 Most of the other nymphs have the right hand on the hip. Some
 have the right arm back and down, bent or straight, but it is
 questionable whether this pose is significantly different
 from hand-on-hip. The extreme case is the figure at 06:30 on
 the outer band, Miss (or Master?) Otarar (dressed, standing
 on an horizontal tube); the first of four dressed figures.
 Miss Ofchdamy, the first of the five "extra" nymphs at the top,
 may be another significant exception, but her
 forearm is not visible on my copy.

f72r3 Cancer

The outer nymph at 11:00, miss Otchy(?)-Daiin, has the right
arm stretched back and down at 45 degrees. She may well be the
leader of that band; there is a wide gap between her and the 
preceding nymph at 09:30.

I cannot see any other nymph with stretched right arm, but 
half of the nymphs are just faint blurs on my copy. 

f72v3 Leo

I see two ladies with the right arm stretched back and down at
45 degrees, bot on the inner band: Miss Oky at 11:30,
and Miss Oteeod(?) at 06:15.  

There is no obvious starting point, but the diagram is 
cut by multiple creases between 07:00 and 10:30, which seems
a natural place to start.

f72v2 Virgo

This seems to be a very complicated month astronomically 8-)
There are many nymphs in new and strange poses, and even a freak
reappearance of the dustbin (shallow, with "cutaway" edge).

I can see several nymphs with the right arm stretched back and
down at 45 degrees.  In the outer band there are Opaiin at 08:30,
and Ofchdy-Sh. at 05:00.  In the inner band we have four consecutive
nymphs starting at 05:00 (Cheosy, Ofcheey, Yteedy, On-Aiin).

However we also have a nymph at 00:15, Miss Oeedy, with *both*
arms stretched back, and hands clasped behind her.  Three 
nymphs (outer Oeedey and Oeeo-Daiin at 10:30-10:45, inner
Oka*** at 10:30) are grasping their stars with both hands;
and inner Okeeom at 01:30 is almost doing the same.

f72v1 Libra

Miss Oteoly at 10:30 on the outer band (the "standard" starting
place), has the right arm stretched back and down. But so do
Miss Okeeoly at 01:00 and Miss Okal at 11:00.

In the inner band the nymphs are holding their right hand in
various positions near the hip; none seems to have a clearly
"stretched-out" arm. The one that comes closest is Miss Oko**y
at 03:30, but she is bending down to avoid the "cigarette
hole", and the hand position my be accidental. In any case
that hole would be a natural starting place for the inner band.

f73r Scorpio

Outer Misses Dolshey and Opaiin at 08:00-09:00 have outstretched
right arms. The latter is more exhuberant and holds a bigger
star.  09:00 could be a starting place in this case.

Ladies Shekal, Okeedy and Okedal at 05:00-06:30 outer band, 
have stretchde arms.  They cannot be all starting points...

In the ineer band, the stretched-arm ladies are Miss Chek and
Miss Kar (not their real names, I am sure 8-) Miss Kar, by the
way, is the one who was involved in the cigarette hole affair
with Taurus girl, as reported bove.

Outside the diagram, at the top, there are Miss Chockhy and 
Miss Yteeody; the latter may a full stop, hardly a start marker...

f73v Sagittarus

I see only three ladies with stretched arms here. In the outer
band we have Miss Ykeody at 02:00 and Miss Okeody at 10:00; the
latter may well be the band leader. In the inner band I see only
Miss Otal at 03:00.

As previously mentioned, I’ve been grinding my way through the Cisiojanus entries for February (mainly from the Usuarium website). However, it turns out that there is far less variation than I supposed. For example, “Ig” (on the 1st February) appears in only a single missal (FR Noyon 1541, Missale Noviomense). And even though St Walburga is listed in some volkskalenders, I haven’t seen “Wal” appear in an single Cisiojanus mnemonic for February 25th. And finally, many of what I first thought were variants are actually nothing more than transcription errors.

Anyway, here’s my current list for February circa 1400-1450:

1Bri / IgSt Brigid of Ireland (v)
2Pur*Purificatio Mariae
3BlaSt Blaise (Blasius)
4Sus
5AgSt Agatha (v)
6Ath / At / DorSt Dorothea (v)
7Fe
8BruBruno of Querfurt, Archbishop in Prussia
9O / ApSt Apollonia (v)
10ScoSt Scholastica (v)
11Las / La
12Ti / Sti
13Ca
14Val / VaSt Valentine (yes, that St Valentine)
15Ent / Lent
16Lu / In / JuSt Juliana of Nicomedia
17Li / Ly
18ConSt Concordia
19Iun / Jun
20Ge
21Tu[n]c
22Pet / Pe*Cathedra Petri (St Peter’s Chair)
23Ru[m] / Tru[m]*[vigilia]
24Mat / Ma*St Matthias, Apostle
25ThiSt Walburga (v)  
26Am
27In
28De / Te

Note that the four starred days are the ones I’ve seen written in red on some (but not all) calendars: most have only one or two red days.

My original plan was to compare these mnemonic syllables with the labels for the Voynich Manuscript’s f70v2. However, this simply doesn’t seem to go anywhere. I’ve tried to line up ot/ok in labels with male saints vs female saints, or with virgin martyrs, but nothing seems to match.

I’ve also looked for (more traditional Art History) subject matches (e.g. St. Apollonia is associated with toothache, St Brigid with protection, blacksmithing, livestock, dairy, etc) in the “nymph” drawings, but so far have found zilch. [The outer ring ‘nymph’ at around 4 o’clock appears to have a beard and no breasts so I’d guess is male, but might he be St Peter? It’s not a very convincing argument, I cannot deny it.]

I’ve also been thinking about this with reference to the Volkskalender B family of manuscripts I discussed here many years ago. For example, BSB Cgm 28, or St Gallen Cod. Sang. 760, or Zurich Ms C. 54 [which has a Cisiojanus column], Pal. Germ. 291, and so on. But this too feels like it’s a busted flush: computus aside (calculating Easter), there’s really not a lot to work with there, calendar-wise.

What’s Left?

My “Attack the Fish!” post mentioned Fribourg Ms L. 309, which inspired commenter Peter Moesli to look beyond the calendar page in that manuscript. He found a health tip for February there: “Beware of the cold and do not wash your head or cut your hair“.

All I feel I currently have left is wondering whether this health tip is broadly the type of ‘secret’ we should be looking for as a possible text match for the rings of the zodiac pages. The Pisces page, for example, has three circular rings, which would amount to roughly 5/6 lines of text. So if these rings are where the actual ‘payload’ is located, it’s surely not a very big payload.

Bah.

In a comment to my last post, I remarked that I suspect the labels on the pages with the zodiac roundels might be verbosely-enciphered syllables of a cisiojanus mnemonic. As an example of how this might work, I expanded out “cisiojanus epi sibi vendicat”: unfortunately, this was for January, and January was (circa 1400-1450) associated with Aquarius, which – along with Capricorn – is one of the two zodiac signs that were (probably) removed from the Voynich Manuscript, alas.

Consequently I need to pick a different target. Hence my plan is to Attack The Fish! (i.e. Pisces, f70v2)

So, what I’ll be doing over the next few days / weeks (though I sincerely hope not months) is exploring the world of Cisiojanus through a fishy February lens. This will involve understanding the saints, syllables, languages, traditions etc. For example, February 1st celebrates St Brigide v. [“v.” = virginis], February 2nd is Purificatio Marie (this is Candlemas, celebrated 40 days after Christmas, which is the day Mary went to the Temple in Jerusalem to be purified, following Jewish law), 5th is for St Agatha, 9th for St Appolonia, and so on.

For anyone wanting to come along for the ride, here’s a link to February from (arguably) the nearest example of an astronomical calendar we have, i.e. the “Astronomical medical calendar in German (Studio of Diebolt Lauber at Hagenau, about 1430 – 1450)“. I haven’t yet read Rosy Schilling’s work on this, but I plan to do this very shortly. Similarly, here’s a cisiojanus column in a volkskalender February from Fribourg MS L. 309, another nice manuscript I’ve previously linked to here. There will be more, many more!

Just so you know, it’s easy to Google web pages with February cisiojanus manuscripts. This is because they typically start “bri pur blasus…” or “bri pur bla sus”. One exceptional source for liturgical calendar stuff is the Usuarium (here’s its page for February), which I hadn’t seen before today.

One of the enduring mysteries of the Voynich Manuscript’s enigmatic “Voynichese” script is that it varies. Not content with having two full-blown ‘languages’/’dialects’ (known as Currier A and Currier B), the way Voynichese ‘behaves’ on a glyph-level, word-level, line-level, paragraph-level, page-level and even section-level varies in many, many other ways (e.g. LAAFU etc).

This pervasive variability is an easy spanner to throw in the works of the kind of ‘simple [universal] explanation’ that gets periodically churned up for Voynichese – you know, mirrored High German, etc. No natural language structure could explain this variability – languages are stubbornly historical and offer students challenges on many levels, for sure, but Voynichese is just something else.

One of the many Voynichese variations is “labelese”: these are short words or phrases that appear to be attached to labels on complicated diagrams (normally with astronomical or zodiacal content, as per the dominant interpretation). Complicating the issue is that labelese can be juxtaposed with non-labelese, such as with this image from the “Aries” (f71r) page:

Here you can see some normal-looking (continuous) Voynichese (on the curved lines of text), together with some shorter labels (in EVA: okldam, oteoaldy, and oteolar). And you might possibly speculate from this that the continuous-looking Voynichese might be normal Voynichese language, and that the labelese might be some kind of simplified subset of Voynichese.

The reason for suspecting a subset is that you almost never see labelese words starting with (EVA) qo- (which is a hugely popular pattern in Currier B – so almost no qok- words or qot- words). And you often see l- initial words in some Currier B (particularly in Q13), which we can see here (but never in Currier A).

But… look again at the image. The outer band of text here runs “okeeedy oky eey okeodar okeoky oteody oto otol oteeyar“. And the inner band of text runs “ockhchy oteesaey lcheotey okarody shs“. The way that so many of these words (both in the labels and the continuous curved text) start with ok- / ot- is more than a bit suspicious, hein?

In fact, you might go so far as to suggest that labelese’s ok- words seem to broadly correspond to (say) Q20’s qok- words (I think that the labelese ok-/ot- ratio is about the same as the Q20 qok-/qot- ratio, but I haven’t checked).

In Q20, y- initial words tend to be the first words in a line (suggesting that there is some clever bastard trickery going on with line-initial glyphs), but in labelese we see very few. Perhaps labelese might be trying to disguise where each individual set of labels start?

So my thought for today is simply this: might ok- be a kind of labelese-specific null? In which case the original text might have been continuous text, which was then divided into small label-sized chunks and had an ok- prepended to most of them.

Darrell Huff’s (1954) “How to Lie With Statistics” is a twentieth century classic that’s well worth reading (I have a well-thumbed copy on my bookshelf that I bought back in the 1980s). It’s basically a breezy introduction to statistics, that concentrates largely on how people get things wrong in order to get across the general idea of how you might (possibly, hopefully) try to get things right in your own work.

A journalist rather than an academic statistician, Huff’s book ended up selling more than 1.5 million copies. You can hear echoes of his reversed-expectations presentation in numerous other book titles, such as Bill Hartston’s “How to Cheat at Chess”.

Sadly, The Truth Is Much, Much Worse

When later I did statistics modules at University, the awful truth slowly dawned on me: even though tools (such as Excel) make it easy to perform statistical procedures, stats really isn’t just a matter of “running the numbers”, cranking out an answer, and drawing some persuasive-looking graphs.

Even just conceiving a statistical experiment (e.g. something that’s based on good data, and that stands a chance of yielding meaningful results) is extraordinarily hard. Designing statistical experiments (e.g. understanding the sampling biases that are inevitably embedded in the data, and then working out how to work around them) is also hugely tricky. Executing them is no mean feat either: and then – finallyinterpreting them is fraught with difficulty.

In general, my own experience of statistical experiments is that at least half are fatally misconceived; of the remainder, half are horribly misdesigned; of the remainder of that, at least half are sadly misexecuted; and of the remainder of that, at last half of the results are tragically misinterpreted. Note that the overall success rate (<5%) is for people who broadly know what they’re doing, never mind idiots playing with Excel.

A Story About Stats

Back when I was doing my MBA, one of the final marked pieces was for the statistics module. When I took a look at the data, it quickly became clear that while most of the columns were real, one in particular had been faked up. And so I wrote up my answer saying – in a meta kind of way – that because that (fake) column was basically synthetic, you couldn’t draw reliable conclusions from it. And so the best you could do in practice was to draw conclusions from the other non-synthetic columns.

I failed the module.

So, I made an appointment with the lecturer who marked it, who also happened to be the Dean of the Business School.

  • I said: Why did you fail this piece?
  • He said: Because you didn’t get the right answer.
  • I said: But the column for the ‘right’ answer is fake.
  • He said: I don’t think so.
  • I said: Well, look at this [and showed him exactly how it had been faked]
  • He said: Oh… OK. I didn’t know that. But… it doesn’t matter.
  • I said: errrm… sorry?
  • He said: you’ve got a Distinction anyway, so there’s no point me changing this mark

And so I still failed the statistics module.

The Voynich Manuscript and Stats

If you think Voynich Manuscript researchers who run statistical tests on Voynichese are somehow immune to these fundamental hazards, I don’t really think you’re paying enough attention.

Until you accept that the core problems inherent in Voynichese transcriptions – there are many, and they run deep – will inevitably permeate all your analyses, you really are just running the numbers for fun.

The main things that bother me (though doubtless there are others that I can’t think of right now):

  • Transcription assumptions
  • Transcription error rates
  • Running tests on the whole Voynich Manuscript, rather than on sections (e.g. Q13, Q20, Herbal-A)
  • How Voynichese should be parsed into tokens (this has bugged me for 20 years!)
  • Copying errors and Voynichese “weirdoes”
  • The bifolios being out of order
  • Whether there is a uniform ‘system’ underlying both Currier A and Currier B
  • The problems with top-line text
  • The problems with line-initial letters
  • The problems with line-final letters
  • etc

With so many parallel things to consider, I honestly think it should be no surprise that most attempts at Voynich analysis fail to achieve anything of value.

Voynich Theories

I have no doubt that researchers do their best to be rational and sensible, but many Voynich theories – or, perhaps more accurately, Voynich ‘approaches’ – are built upon a fundamentally flawed statistical ‘take’, e.g. that Voynichese is just a simple (but highly obscure) text.

Unpopularly, this seems to be true of just about all ‘Baxian’ Voynich linguistic analyses. Statistically, nothing supports the basic assumption of a ‘flat’ (but obscure) language. In fact, Voynichese is full of confounding, arbitrary, difficult, unlanguagelike behaviours (see the incomplete list above), all of which you have to compensate for to get your data to a point where you even begin to have something remotely language-like to work with. But hardly anybody ever does that, because it’s too tricky, and they’re not genuinely invested enough to do the ‘hard yards’.

It’s also true of Gordon Rugg’s table ‘take’; and of just about all simple ciphers; and – also unpopularly – of hoax theories (why should meaningless text be so confounded?) And so forth.

The sad reality is that most researchers seem to approach Voynichese with a pre-existing emotional answer in mind, which they then true to justify using imperfect statistical experiments. More broadly, this is how a lot of flawed statistical studies also work, particularly in economics.

In fact, statistics has become a tool that a lot of people use to try to support the lies they tell themselves, as well as the lies their paymasters want to be told. This is every bit as true of Big Oil and alt.right politics as of Voynichology. Perhaps it’s time for an even more ironic 21st century update to Darrell Huff’s book – “How To Lie To Yourself With Statistics”?

From the time of writing “The Curse of the Voynich” to now, I’ve felt annoyed with myself for not being able to figure out precisely what is going on with the Voynichese “daiin daiin” pattern. Back in 2006, I wondered whether all the scribal variations of EVA ain in Herbal A might actually be somehow enciphering Arabic numerals. Sit back with a drink and I’ll tell you the story about how I got to that point…

My 2006 Voynich hajj

While preparing to write Curse, I went to New Haven on a Voynich research hajj. The Beinecke curators had been kind enough to allow me to spend a few days looking at every page in depth, though I must confess that I left the nine-rosette page until last because I was frankly terrified it would fall apart in my hands. The hunch I had that I should look at the marginalia with a black (UV) lamp had paid off, because it revealed the (faded away) Voynichese at the top of f17r.

So, after a couple of days of doing this, I had already gone through just about every folio: but I kept having a nagging feeling that there was so much of it to look at that I had missed something really big (and simple), hidden in plain sight. I therefore decided to pick a single early page and just stare at that for an hour, to try to ‘go deep and narrow’ (rather than wide and shallow).

I picked f38v (because the vellum was clean and the writing was clear), whose text looks like this (interrupted by a vertical plant stem):

After more than half an hour of staring from different sides and angles, what suddenly struck me was that a surprising number of the aiin-group words appeared to have had their terminating scribal loop added on as a second pass. If you extract out the aiin-family groups from this page, you get this:

In Curse, I subsequently speculated that the original form of these shapes might have looked like a more conventionally-written “aiiv”, but with a dot placed above (or possibly in-between) one of the three ‘peaks’ of the ‘m’-shape (i.e. if you read it as “am”); and that a second writing pass might have turned this dot into the starting point for a downward scribal loop (i.e. to conceal the location of the dot).

Of course, nobody then or now agreed/agrees with this speculation (which is OK). All the same, I know what I saw back then, and I have to say that the scans don’t properly capture what I saw. As always, be careful that a map isn’t the territory (not even for Borges), and a scan of an artifact isn’t the artifact.

EVA and aiin scribal loops

Even if you don’t agree with my 2006 speculation, from this page alone there would seem to be a wide range of scribal forms used when writing the terminating loop of the EVA “n”, ranging from really short to really quite long. Simply assuming that these ‘can only be’ scribal loops would therefore seem to be quite a foolish first step. Unfortunately, this is exactly what the EVA designers did.

I can see exactly why they did it (essentially, they were trying to design a transcription alphabet to enable a kind of interpretation-free scholarly discourse): but I think it would be wise to bear in mind that they might just have oversimplified things as far as daiin goes.

Glen Claston’s Voynich 101 transcription arguably went too far in the other direction, and was perhaps wrapped too tightly around the specific way (based on Leonell Strong’s reading) he parsed Voynichese, which I think was not a good choice for quite different reasons. But… it is what it is, as always. Caveat lector, for sure (and not just if you are Clarice Starling).

All the same, I personally can’t help but be suspicious of the argument that Lisa Fagin Davis tried to project onto the aiin groups, that their terminating loops are a scribal ‘tell’ universal across the pages of the Voynich Manuscript. For Currier B pages, this is perhaps true: but I don’t appear to see the same consistency in Herbal A pages. Sure, use EVA k and EVA t as palaeographic tells all you like: but please be careful trying to draw the same category of conclusion about scribal flourishes in Herbal A pages, you might just be throwing the cryptographic baby out with the palaeographic bathwater.

daiin-family repetitions

Lastly, there’s something acutely uncomfortable about some of the daiin-family repetitions. f38v has a fine example of this, in that it has five of them in a row. Here are the first four (with the stem in the middle):

…and here’s the next one along…

In fact, the next few words are all very daiin-heavy:

This is a big cluster of daiins, which I can’t help but wonder might be days in a date, such as “1440”. Though in the 15th century, I should add that it was also common to omit the initial “1” in dates, i.e. “440”.

Could it be that “daiin-daiin” as a pair might somehow encode a single Arabic numeral, i.e. even more verbosely than you might imagine? I don’t know, but I thought I’d raise this as a possibility.

If you’ve previously looked at the final page of the Voynich Manuscript (f116v), you probably know exactly what I’m talking about here: the strange marginalia / pen-trials / whatever that aren’t Voynichese, but aren’t obviously anything else either. The middle two lines of “michitonese” are interspersed with crosses, making them resemble a prayer or protective charm (when reading these out, you’re supposed to make the sign of the cross when you see a ‘+’). The two michitonese words I’m going to focus on today look like “ahia maria“, which is what “Saint Mary” would sound like in Greek. But is that what they really say?

Parsing the page

To my mind, the initial challenge is working out not how to read the page (Lord knows we don’t need yet more speculative michitonese decipherments), but rather how to parse the page. Knowing that crosses are a structural feature of prayers and charms is a helpful start, but I suspect that differences in inks and letter shapes are also key things to bear in mind.

I discussed this page in The Curse of The Voynich (2006), and proposed that what we are often seeing here is a page whose writing had faded so much by the time that a later (in fact, probably much later) owner looked at it that they felt compelled to emend it before it disappeared completely from sight.

Along those same line, I’ve previously wondered whether the first two words of the pair of lines (which have been immortalised as “michiton oladabas”) might originally have read “nichil obstat”, before fading almost to nothing and being emended into their current state. This isn’t just a matter of ink density, as some people have suggested: for example, the ‘l’ of oladabas looks like it was written by a 15th century quill pen but the ‘n’ of michiton looks to me like it was written by a 16th century metal nib. And if you compare all three ‘a’s of oladabas, you’ll struggle to work out which is original and which is emended. (For what it’s worth, I believe the middle ‘a’ is original and the other two are emendations.)

So broadly speaking, I feel that the first thing we have to do is parse “ahia maria”, i.e. try to work out which parts of it are original and which are likely later emendations.

Multispectral scans

Though we only have (frankly shocking) quality multispectral scans of f116v that were left on the Beinecke’s file server several years ago, these are vaguely helpful here. Here’s what I was able to salvage from “Voynich_116v_WBRBB47_108_F.jpg”:

Though this is hard to make out, it suggests that the final ‘a’ of (what looks like) “ahia” wasn’t emended, but the rest of the word was. The first “ma” of “ma+ria” also seem to have been worked on.

In general, it seems that we can probably trust that the “ma+ria” part was what was originally written, but we need a bit more insight before we can come to a conclusion about the “ahia” part.

The History of Hail Mary

Given the prayer-/charm-like context, it would seem very reasonable to hypothesise that the two words might originally have read Hail Mary (in some language). But when did people start saying Hail Mary?

The Greek form (which begins Θεοτόκε Παρθένε, χαῖρε, κεχαριτωμένη Μαρία, ὁ Κύριος μετὰ σοῦ. ) developed independently of the Latin form “Ave Maria, gratia plena“. The latter took shape around the turn of the first Millennium, with Thomas Aquinas noting that the biblical original (“Ave, gratia plena“, from the Gospel of Luke) had had the word “Maria” inserted. The Latin form first appeared in print in 1495 in Girolamo Savonarola’s “Esposizione sopra l’Ave Maria“.

In terms of what forms of Hail Mary we might reasonably be looking to find, then, the Greek version is distinctly unpromising, while the Latin is very much limited to Ave Maria.

Yet if you search for medieval Latin charms that mention Mary, you’ll find a good number that invoke her name (e.g. “+ sancta Maria peperit +” in CUL Additional 9308, folios 49r-50r, mentioned here, footnote 155), often when hoping to ensure a good outcome from childbirth.

Additionally, “Maria Maria” appears doubled in a particular branch of Gaudendum nobis est (see MSS A-GU 30, CH-EN 1003, CH-EN 102, and Stuttgart 95, mentioned here, Table 4-6, which are from Engelberg and Lambrecht). We’ll see why this is relevant in the next section…

The Other Maria(s)

We have to be careful that we don’t let people’s prior readings over-influence our attempt to make sense of the page. The “six + marix + morix +” text on the same line is visually suspicious to me, simply because both the marix and the morix look a lot like maria, which all matches up with the rest of the line:

That is, the line could easily, when taken as a whole, be saying something like “six + maria + maria + vix + ahia + ma+ria +”, i.e. very much less than you might at first think.

Multispectrally, the morix shows signs that its ‘o’ and ‘x’ were emended, but the rest left intact:

Of course, “six” isn’t actually a Latin word, so that too is an interpretative reading of the line. Moreover, its middle “i” is shaped completely unlike the ‘i’s in any of the Mari[ax] words, so was almost certainly emended. So: if some of the original text’s a’s were later misread (and emended) as ‘x’s, might this have originally been an abbreviation for “Sancta”, and so read more like “S[anct]a + Maria + Maria + vix + Ahia + Ma+ria +”?

And again, “vix” (= hardly, scarcely, barely, only just, with difficulty, reluctantly, etc) is perhaps not a Latin word you might expect to see in a charm or prayer: its middle “i” is shaped completely unlike the ‘i’s in any of the Mari[ax] words, so also looks like an emendation. So this too is a bit of an awkward fit.

Your thoughts, Nick?

Well, I wasn’t clear on this in 2006, and to be honest I’m not much further along in 2023. It seems to me that we’re looking at this page through the wrong kind of technical lens, because we’re not really able to answer even the simplest of questions about its codicology. Multispectral imaging was a bit of a bust for f116v, though to be fair this is exactly what a lot of the literature points out for studying pages with different ink layers (you really need Raman imaging to separate out different inks).

A different approach, which I’ve proposed for a while now, would be to map the surface of f116v to see if we can detect differences in the depth of pen indentations between different codicological layers, and then pick out the earliest layer for its meaning. And in fact, in a post on BBC Futures today, this is exactly what the Factum Foundation in Madrid has been doing with the Archiox Project for the Bodleian library.

Who knows what intrepid codicologists might find hidden in the surface of the Voynich Manuscript?

Like most people, I’ve read numerous articles about ChatGPT recently. Perhaps the most entertaining was where ChatGPT (sort of) played chess against the #1 chess engine Stockfish, with hilarious consequences (if you’re a chess player, that is).

However, few articles manage to get the point across that ChatGPT is not an all-knowing Oracle, but rather something that is trying to please you by providing answers that are as realistic-looking as it can make them (rather than actually realistic). It’s entertainment, people (and definitely not real).

Perhaps if I gave some examples you’ll see what I’m talking about…

“Did Filarete have his own herbal manuscript?”

I asked ChatGPT this question, because it was something which I already knew the answer to (and had blogged about back in 2008, and also since). The right answer is that Filarete did have his own herbal manuscript (it was mentioned in fol.78r of MS Ashburnham 198, “Sed et Antonius Averlinus Philaretus lingua vernacula scripsit eleganter“), but as far as I know, nobody knows if it or a copy of it still exists.

ChatGPT started safely and correctly:

There is no known herbal manuscript that can be definitively attributed to Filarete, the fifteenth-century Italian architect and writer. Filarete is best known for his architectural treatise, the “Treatise on Architecture,” which he wrote in the early 1460s.

ChatGPT

However, our chatty digital friend then went a bit off-piste:

However, there are some herbal manuscripts from the same general period that are associated with Filarete or have been attributed to him. For example, there is a manuscript in the Bibliotheca Augusta in Wolfenbüttel, Germany, that is sometimes referred to as the “Herbal of Filarete.” This manuscript contains descriptions and illustrations of various plants and their medicinal uses, but it is not clear whether Filarete was actually the author of the manuscript or was simply associated with it in some way.

ChatGPT

Interesting, I thought, even if I didn’t believe a word of it. It went on:

There are also other herbal manuscripts from the fifteenth century that are similar in style and content to the “Herbal of Filarete” and may have been produced by the same workshop or circle of artists and writers. These manuscripts are sometimes referred to as the “Herbarium of Apuleius” or the “Herbarium of Pseudo-Apuleius” and contain similar descriptions and illustrations of plants and their uses.

ChatGPT

“What is the shelfmark of that “Herbal of Filarete” you mentioned?”

I still didn’t believe a word, so asked it: “What is the shelfmark of that “Herbal of Filarete” you mentioned?” The response I got back was a little unexpected:

I apologize for the confusion in my previous response. Upon further research, I could not find any evidence of a manuscript called the “Herbal of Filarete” in the Bibliotheca Augusta in Wolfenbüttel, Germany.

I believe there may have been a misunderstanding or confusion with other sources. While there are many herbal manuscripts from the medieval period, there is no known herbal manuscript that can be definitively attributed to Filarete.

I apologize for any confusion or misinformation my previous response may have caused. If you have any further questions, please feel free to ask.

ChatGPT

It’s certainly true that the Bibliotheca Augusta in Wolfenbüttel has at least one Pseudo-Apuleius manuscript, so it looked like there was just some confusion going on. Gigantic input dataset aside, ChatGPT really isn’t anything like Google, so it would be grossly unfair to imagine that it is doing anything like the same thing when it tries to answer a query.

So I decided to take a different tack, to see if I could flush anything out about what was going on.

“Are there any herbal manuscripts associated with Filarete?”

Maybe there’s a herbal manuscript out there that’s loosely associated with Filarete or his coterie that I don’t know about, I thought. So I tried to poke the bear again (but in a slightly gentler way than before).

Can you guess how ChatGPT responded? I think you can. As before, the first paragraph was fairly solid:

To my knowledge, there are no herbal manuscripts that can be definitively attributed to Filarete, the fifteenth-century Italian architect and writer. Filarete is best known for his architectural treatise, the “Treatise on Architecture,” which he wrote in the early 1460s.

ChatGPT

But you might find the next paragraph eerily familiar:

While Filarete is not known to have authored any herbal manuscripts, there are some medieval manuscripts that contain descriptions and illustrations of plants and their medicinal uses that have been associated with him or with his circle of artists and writers. One example is a manuscript in the Biblioteca Medicea Laurenziana in Florence, Italy, that is sometimes referred to as the “Herbarium of Filarete.” This manuscript contains descriptions and illustrations of various plants and their medicinal uses, but it is not clear whether Filarete was actually the author of the manuscript or was simply associated with it in some way.

ChatGPT

Once again, this seems like an historical hallucination on ChatGPT’s part. The odd part is that it reprised the exact shape of its previous hallucinatory answer, but referring to a different top-end library. So, I again asked for the shelfmark of this imaginary manuscript…

“What is the shelfmark of that manuscript in Florence?”

That’s what I asked: and, sadly, you already know what ChatGPT’s reply is going to look like, because it’s basically the same as before:

I apologize for the error in my previous response. After further research, I have found that there is no known “Herbarium of Filarete” manuscript in the Biblioteca Medicea Laurenziana in Florence, Italy.

While Filarete is not known to have authored any herbal manuscripts, there are some medieval manuscripts that contain descriptions and illustrations of plants and their medicinal uses that have been associated with him or with his circle of artists and writers. One example is a manuscript known as the “Herbarium of Apuleius,” which is sometimes referred to as the “Herbal of Pseudo-Apuleius.” This manuscript is kept at the Bibliotheca Augusta in Wolfenbüttel, Germany, and has the shelfmark Codex Guelf. 82.7 Aug. 2°.

ChatGPT

What to make of this?

ChatGPT is undoubtedly a brilliant technical achievement. But as far as I can see, for now it’s just part of the entertainment industry.

Here, I do wonder whether the format of the question I asked kind of loosely defined the format of the answer ChatGPT inferred I was hoping to receive back: and that it then hallucinated enough additional stuff to complete the gaps in the answer format to provide a well-formed-looking answer. So maybe the whole point of ChatGPT is that the shape of the question defines the shape of the answer, even if there aren’t any actual facts inside the answer. Something to think about, anyway!

Over at voynich.ninja, there’s an interesting recent thread on the (in-)homogeneity of Voynichese, i.e. how consistent (or inconsistent) with itself the Voynich Manuscript’s text is (either locally or globally). Given that I have been working on Q20 recently, I thought it might be interesting to take a brief look at that quire through this particular lens.

In Search of a Problem Statement

One intriguing side of Voynichese is that even though it exhibits high-level consistency (e.g. the continuous script, plus the well-known differences between Currier A pages and Currier B pages), medium-level consistency (e.g. thematic-looking sections such as Q13, Q20, Herbal-A, Herbal-B etc), and even bifolio-level consistency (more on this below), there are open questions about the apparent lack of low-level consistency.

In particular, Voynichese ‘words’ (which have been the subject of countless studies and analyses) present many apparent local inconsistencies. As Torsten Timm pointed out in the voynich.ninja thread referenced above, words that are extremely common on one page of a section can be completely absent from the next. And, awkwardly, this is sometimes even true for pages that are the recto and verso sides of the same folio.

Even though there are countless ways to airily explain away these kinds of inconsistencies (change of subject matter, change of source structure, change of underlying plaintext language, change of local cipher key, etc), all too often I think these are invoked more as a research excuse for not actually going down the rabbit hole. (And I for one am bored stiff of such research excuses.)

So, before we start reaching gleefully for such cop-out answers, we need to first properly lock down what the core low-level consistency problem actually is. Basically, what specific behaviours can we point to that indicate that Voynichese has a problem here?

Captain “ed”

It was WWII codebreaker Captain Prescott Currier himself who pointed out nearly fifty years ago that you could usually tell Currier A pages from Currier B pages simply by looking at the proportion of ‘ed’ glyph pairs on that page. (Currier A pages have almost none, Currier B pages normally have loads.)

Personally, I’d add some caveats, though:

  • Even though it might be tempting to think of “ed” as a bigram (i.e. a single token), it seems far more likely to be a contact boundary between an “e”-family token (i.e. e/ee/eee) and a “d” glyph.
  • To me, there often seems to be something funny going on with qokedy / qokeedy / etc that isn’t really captured by just looking for “ed”

Helpfully, you can use voynichese.com’s layer feature to bring to life the variation in Voynichese words containing “ed”, e.g. this query for lots of different subgroups of “ed” words. Even though Herbal A pages have basically no ed pairs at all, the ed’s nothing short of explode at the very start of Q13:

The first three pages of Q20 are very nearly as colourful:

For ed, it seems to be the case that recto and folio pages have a similar kind of ed-density: for example, if you compare f107r/f107v with f108r/f108v, you can see clearly that the two halves of each folio seem quite similar:

The f111r/f111v pair seems to buck this trend slightly, insofar as f111v (on the right) seems somewhat less ed-dense than its recto side f111r (on the left):

While I’m here, I’d note that f116r (the last proper Voynichese page of Q20) seems to have a structure break halfway down, which would be consistent with an explicit and/or a colophon placed at the end of a chapter / book:

There’s also the question of whether the two folios making up each bifolio appear ed-consistent. I’d say that this appears true for most Q20 bifolios (e.g. f103 and the top half of f116r, f104-f115, f105-f114, f106-f113) but certainly not for others (e.g. f107-f112 and f108-f111). It’s very hard to be definitive about this.

Finally, I’d also note that while Quire 8’s f58r/f58v (with their starred paragraphs) do have some ed-words, their ed-fractions are extremely low, which would make classifying them as “pure” Currier B difficult:

Torsten Timm’s “in”

Torsten Timm has similarly looked at what the usage of the Voynichese glyph pair “in” tells us. Of my own set of voynichese.com experiments, the one that seemed to me to be the most interesting was comparing “iin” with “[anything else]in”.

For example, even though iin dominates [^i]in for most of the Voynich Manuscript, the first folio of Q13 has almost no “iin”s in it at all:

Folio f111 is also a little bit odd, in that its verso side has many more [^i]in words:

“ho”-words Way

As with Currier’s “ed”, “ho” is very much a contact locus between two families of glyphs: on the left, you have ch/sh/ckh/cth/cph/cfh, while on the right you have or/ol/ok/ot/op/of etc. As such, it looks like a useful way of exploring for a group of glyph boundaries, but this does need to be carefully qualified.

If we visually group this ho-transition (via voynichese.com) in terms of the origin of the “h”, we get a query that looks like this. This reveals that most ho instances are in fact “cho” (dark blue). However, the f93r/f93v folio does look particularly unusual in this respect:

The final two paragraphs of f116r are also unusual, this time for their almost complete lack of ho-words:

If you try to classify ho-words in terms of what follows, you seem to get less predictability.

Putting ed / in / ho Together

From the preceding sections, I’d say that the overwhelming impression I get is that pages within a folio (and indeed pages within a bifolio, though to a slightly lesser extent) are actually reasonably consistent with each other, and with relatively few counter-examples.

Unsurprisingly, this is also what we see if we simply merge the three ed / in / ho queries into a single voynichese.com query. Here, we can easily pick out the dishonourable exceptions, such as f111 (where f111r is dominated by “ed” [blue], yet where f111v is dominated by “in” [red]):

If we instead highlight cho and sho separately, what emerges is that, unlike the rest of Q20, the f106-f113 bifolio has a surprisingly high proportion of sho-words (in yellow):

I could go on, but I think my visual argument here has pretty much run its course.

Thoughts, Nick?

Even though Torsten Timm used ed / in / ho as part of his argument concluding that Voynichese pages are independent of each other, I’m not sure I fully accept his conclusions. (He’s certainly right about words, but the details and ramifications of that are for another post entirely.)

For me, the behaviour of ed / in / ho seems to suggest something arguably even more unsettling: which is that there seems to be consistency at the bifolio level.

And so it seems that we’re facing a BAAFU (“Bifolio As A Functional Unit”) scenario here. Which is arguably even more mysterious than Currier’s LAAFU (“Line As A Functional Unit”), wouldn’t you agree?