22-04-2022, 07:34 PM
I support anyone who is interested in "solving" the text to perform the basic frequency analyses like Searcher. Nothing is more convincing than looking at the numbers yourself and realizing the issues that are being faced in getting Voynichese (at the glyph level) to look and act like natural language.
Maybe this explanation will help for understanding the issue better. This approach is far, far from being unique to me, but it helps me keep on track when trying to come up with ideas to test.
Voynichese is incredibly contradictory in that depending on the level of examination the statistical results look like something that could be a natural language. However, these initial impressions do not hold when moving from level to level. The place where Voynichese becomes most dysfunctional, in my opinion, is at the glyph level. Of course, there is controversy at what constitutes "a glyph" and that is where all the various transcription approaches come into play. But that issue aside, it remains that the text does not act like any historically feasible natural language to allow an association between particular glyphs (no matter how they are defined) and how alphabetic letters are used in such natural languages.
One major issue is the low character entropy exhibited by Voynichese. Having one glyph in hand, it is simply too predictable what the next glyph is. Searcher described it a different way, but it is just another way of saying the same issue -- certain bigrams (and trigrams) are enormously over-represented if the text is to be seen as a natural language. In my opinion, the clearest conclusion from this result is that it is unlikely to the point of impossible for Voynichese to be a 1:1 substitution no matter how many languages/dialects are looked at.
So "something else" or, almost certainly at this point, a number of "something elses" have to be going on, if you are going to hold with the position that Voynichese is a repeatable cipher process with a clear text underneath.
Koen's work isolated one possible "something else" approach. Can collapsing certain bigrams and/or trigrams into single entities "fix" the very low character entropy? And in particular, which particular bigrams and trigrams give the best "fix" to the low character entropy? This is a complex problem that Koen had to act like a computer optimization program to answer. RobGea has subsequently (in this thread) confirmed his results using more traditional computer testing.
The resulting list of bigrams and trigrams is the list I am talking about. Because I really wanted to solve the entropy issue and I like the idea of a verbose cipher, I have used this list (in combination with other cipher approaches -- namely some historical verbose ciphers) to try to come up with a way to encode Latin and German, specifically Bavarian, into Voynichese. I started with Latin and German because you have to make choices when you do tests and these seemed feasible to me because of other characteristics of the manuscript (e.g. likely image sources, etc).
But it remains that to date, I (like everyone else) am unable to find a group of approaches that allows for consistent (reproducible) cipherization of either of these languages into Voynichese. In other words, there are other aspects of Voynichese that are not solved by the combination of these approaches alone, with the assumptions I have made. Please note that the problems that remain are pretty large and just switching to a different source language isn't going to "fix" these issues. High among these issues is the issue of the Currier A and B "languages."
That being said, I, too, am quite interested in what Rene has found and hope he will be ready to share his approach at some point -- maybe at the conference?
Maybe this explanation will help for understanding the issue better. This approach is far, far from being unique to me, but it helps me keep on track when trying to come up with ideas to test.
Voynichese is incredibly contradictory in that depending on the level of examination the statistical results look like something that could be a natural language. However, these initial impressions do not hold when moving from level to level. The place where Voynichese becomes most dysfunctional, in my opinion, is at the glyph level. Of course, there is controversy at what constitutes "a glyph" and that is where all the various transcription approaches come into play. But that issue aside, it remains that the text does not act like any historically feasible natural language to allow an association between particular glyphs (no matter how they are defined) and how alphabetic letters are used in such natural languages.
One major issue is the low character entropy exhibited by Voynichese. Having one glyph in hand, it is simply too predictable what the next glyph is. Searcher described it a different way, but it is just another way of saying the same issue -- certain bigrams (and trigrams) are enormously over-represented if the text is to be seen as a natural language. In my opinion, the clearest conclusion from this result is that it is unlikely to the point of impossible for Voynichese to be a 1:1 substitution no matter how many languages/dialects are looked at.
So "something else" or, almost certainly at this point, a number of "something elses" have to be going on, if you are going to hold with the position that Voynichese is a repeatable cipher process with a clear text underneath.
Koen's work isolated one possible "something else" approach. Can collapsing certain bigrams and/or trigrams into single entities "fix" the very low character entropy? And in particular, which particular bigrams and trigrams give the best "fix" to the low character entropy? This is a complex problem that Koen had to act like a computer optimization program to answer. RobGea has subsequently (in this thread) confirmed his results using more traditional computer testing.
The resulting list of bigrams and trigrams is the list I am talking about. Because I really wanted to solve the entropy issue and I like the idea of a verbose cipher, I have used this list (in combination with other cipher approaches -- namely some historical verbose ciphers) to try to come up with a way to encode Latin and German, specifically Bavarian, into Voynichese. I started with Latin and German because you have to make choices when you do tests and these seemed feasible to me because of other characteristics of the manuscript (e.g. likely image sources, etc).
But it remains that to date, I (like everyone else) am unable to find a group of approaches that allows for consistent (reproducible) cipherization of either of these languages into Voynichese. In other words, there are other aspects of Voynichese that are not solved by the combination of these approaches alone, with the assumptions I have made. Please note that the problems that remain are pretty large and just switching to a different source language isn't going to "fix" these issues. High among these issues is the issue of the Currier A and B "languages."
That being said, I, too, am quite interested in what Rene has found and hope he will be ready to share his approach at some point -- maybe at the conference?
![[+]](https://www.voynich.ninja/images/collapse_collapsed.png)