Voynich words evolve over a linear sequence

[Mauro]

One more small question please: if I understood correctly, your previous study worked by bifolios, while the one in this thread works by manuscript sections (all the bifolios of each section clumped together), is it right?

[npcompl33t]

My working hypothesis was that text was generated/encrypted using paper volvelles (note the plural). These would inevitably wear out and literally fall apart from repeated rotations, so these would be replaced by new volvelles. These new volvelles would still satisfy the original system, but would not necessarily have the exact same glyphs/combos. Thus, there would be abrupt points where certain words and glyph combos would disappear, appear or change frequency.

I've also considered volvelles in the past, especially after reading Zaterra's 2022 slot grammar paper. Something along the lines of Ramon Llull's Ars Demonstrativa, or perhaps the zā’irja, would produce slot preferences and even supply motive for the construction of a unique script in Llull's case.

In the past when i've tested these I haven't found a combination that could reasonably produce the observed voynich. However I have not revisited these ideas with respect to a changing gradient. I'm not certain if swapping volvelles would cause a gradient like the one the data appears to present, I would imagine they would produce stepwise changes, but that is actually easy enough to test.

---

One more small question please: if I understood correctly, your previous study worked by bifolios, while the one in this thread works by manuscript sections (all the bifolios of each section clumped together), is it right?

That's correct, this groups them altogether by section. You can actually see the bifolio level view of the same data, it looks much noisier than the above section totals. 

---

But one would expect the changes to be "random" only if one expects the text to be gibberish.  I guess that this is your preferred theory, yes?  

@Jorge_Stolfi This is not my position at all, the point i'm making across these posts is that it appears very possible that semantic meaning is preserved over these transformations. That is what I'm trying to test -- is there a convincing way words could pass through this evolution that appears to preserve meaning. For example, the changes might be a result of them altering the script, perhaps adding abbreviated forms or elaborations. That is why I am also testing vulgate. I want to see how a language variant optimized for abbreviations affects the change in word frequencies. 

There are two outcomes: either there appears to be a convincing way that words are being changed that preserves meaning, or the transformations appear to destroy information. The latter would be strong evidence for gibberish, but so far, I have not seen that the former is not possible. 

Which I think is the case for VMS sections like Herbal-A or Stars: at the present stage, I do not see the justification for splitting them into sub-sections.

Even if you don't think there is a gradient, there is ample evidence these exhibit different enough glyph and word patterns to support splitting them into sub-sections.

[Jorge_Stolfi]

I think your example with books is surely valid, but too simplified for the work @npcompl33t is doing.  Having only the heights (one dimension) I agree with you that's impossibile to determine if the varying heights are due to a temporal evolution, but when the dimensions are many (the colour of the covers, the paper quality, the font size etc. etc.) and they co-vary this is a strong indication that there's an underlying evolution.

But, again, there are dozens of attributes of a book that are significantly correlated to its height -- like width, thickness, volume, weight, margin width, letters per page, total text length, number of paragraphs, number of the last page, length of the index, thickness of the cover, ...  And yet the ordering of books by any of these attributes, or any combination of them, will still have no relation to time.

Example: we have books 10 cm high with a blue cover and books 10 cm high but a red cover, then 9 cm high an blue, 9 cm high and green, 8 cm and green and so on. I think it's quite reasonable to conclude that the evolution probably was (10cm, red) -> (10cm, blue) -> (9cm, blue) -> (9cm, green) -> (8cm, green) and so on.

That is not a reasonable conclusion at all.  Publishers do not choose the size of a book or the color of its cover in some strict ordered sequence.  

If indeed you see a correlation between the size of books from some publisher and the color of their covers, it may be simply that many of those books are parts of series with similar size and consistent styles.  Langenscheidt published many "pocket" dictionaries with the same small size and a characteristic bright yellow cover. Springer issued thousands of Lecture Notes in Computer Science with red and silver cover, and many Lecture Notes in Artificial Intelligence with blue and silver covers.  But all those books were published over half a century or more, interspersed with other books with other sizes and colors.  

As you know, I also don't agree with you on the statement that 'the statistics of a text mainly depend on the topic, author, style etc. etc.'. ... Italian texts hundreds of years apart, with different topics and literary genre all clustered together for instance on the bigrams statistics.

Sorry, I was considering texts in the same language.  Which is almost certainly the case of the various sections of the VMS -- including "Language A" vs "Language B."  

if the language is the same, then the topic, style, and author are more significant factors for text statistics than the mere passage of time.  Therefore, the statistical similarity between (say) Bio and Stars cannot be taken as evidence that they were written closer in time to each other than to (say) Herbal-A.  It may well be due to both being about similar topics, and thus using similar words, sentences, verb tenses, etc.  The temporal sequence may well have been Bio-HerbA-Stars or any of the other 6 possible orders. 

the quite borderline List of daily Saints of the Catholic church

But the style of some sections of the VMS could be just as "borderline".  Like the Alchemists Herbal in the original Latin:

• Si esset aliquis qui haberet membrum fractum videlicet ossum. Accipiat de ista herba et pistet cum asongia castroni et empastrentur et ponatur emplastrum super feritam aut fracturam ossis et saldabitur os vel vulnus. Item si femina non posset concipere filios, comedat de foliis istius herbe per spatium XL dierum; subitio concipiet utendo marito suo. Item si quis habet linguam paganam. Accipiat folia istius herbe et confricet linguam superius et inferius et subito curatur. Item si quis habet dolorem propter aliquam decensionem de alto. Accipe de ista herba et fac pulverem et da ei bibere cum bono vino per spatium XV dierum et liberabitur a dolore. Item si quis esset crepatus; utatur de ista herba et subito sanatur.
  

Note the repetion of words like item and accipiat and herba/e/is and the excessive use of et.  And the repetitive structure 

• Item si [someone][condition]. Accipiat ista herba et [procedure] et [resolution]
  

with a certain repertorie of variations.  These style features will skew the word frequencies, which in turn will skew the character and bigram frequencies.  And another section may have a style that is just as peculiar, but in a very different way -- which will bias the statistics in a different direction.

And comparing bigram statistics makes sense only if the spelling is mostly the same.  If you take a text in German with umlauts and the standard spelling, and another text  with ue/ae/oe and where "sch" is written "x" and "ie" is "i" and "ch" is "q" or "shh" depending on being hard or soft, you may well conclude that they are more different among themselves than either is to Basque or Armenian or whatever.

All the best, --stolfi

[npcompl33t]

And comparing bigram statistics makes sense only if the spelling is mostly the same.  If you take a text in German with umlauts and the standard spelling, and another text  with ue/ae/oe and where "sch" is written "x" and "ie" is "i" and "ch" is "q" or "shh" depending on being hard or soft, you may well conclude that they are more different among themselves than either is to Basque or Armenian or whatever.

I suspect you know this is not really true.

But just in case, I happen to have reference texts for all of these, so I tested this statement using the exact same distance metrics I used to calculate my golf statistics. This is a great example, but not in the way you had implied.

   

Not only is the method sensitive enough to detect these differences accurately, it shows the difference within one of my subgroupings is around the same as  the German spelling examples. While the differences between my subgroupings, spans the range of the difference between German and Basque.

[BrianCham1994]

I've also considered volvelles in the past, especially after reading Zaterra's 2022 slot grammar paper. Something along the lines of Ramon Llull's Ars Demonstrativa, or perhaps the zā’irja, would produce slot preferences and even supply motive for the construction of a unique script in Llull's case.

In the past when i've tested these I haven't found a combination that could reasonably produce the observed voynich. However I have not revisited these ideas with respect to a changing gradient. I'm not certain if swapping volvelles would cause a gradient like the one the data appears to present, I would imagine they would produce stepwise changes, but that is actually easy enough to test.

I've been out for a while. Could you please link me to this Zaterra 2022 paper?

I think the mistake of previous volvelle/slot/combinatoric approaches was to assume one generative volvelle, or the entire VMS as a homogenous vocabulary. This generation wouldn't produce enough distinct words to cover the whole manuscript. My idea was a bit more realistic - five simultaneous volvelles, with gradual replacements over time. This creates a slightly different "sub-vocabulary" for each part of the VMS and the total range of words would be expanded compared to a static set of generators.

If it's easy for you to test a stepwise change, I'd be curious to know if you detect such an effect. You are not allowed to view links. Register or Login to view. suggests some combos do appear/disappear quite abruptly, with -oedy being suddenly replaced by -edy even in the middle of the same folio. Perhaps this represents a volvelle being replaced wholly, or maybe just one ring with suffixes. This is only if you're open to it though - no pressure either way. I don't have the capacity to contribute, so one day I'll just dump my old ideas into a new post in case it inspires anyone else. Up to you if you want to think about any of my ideas so far!