The Voynich Ninja

Thanks, that's a very useful overview!

Are the percentages computed over word tokens or word types? I.e.: if qokal appears 10 times, does it count for 10 or for one?

(04-02-2016, 01:06 PM)ReneZ Wrote: You are not allowed to view links. Register or Login to view.Thanks, that's a very useful overview!

Are the percentages computed over word tokens or word types? I.e.: if qokal appears 10 times, does it count for 10 or for one?

The analysis was for the set of words in the text's vocabulary, so a replacement of "k" with "t" in "qokal" would be processed only once.

Thanks! That's the more useful statistic.
I was surprised to see values for the k/t (and other) pairs much lower than expected, but that can then probably be explained.

About (i.e. very roughly) half the vocabulary of the MS are words that occur only once in the entire text.
These are less likely to occur each 'in both versions'.

Job, your statistics are very interesting. For the letter replacements in other languages, were these the highest %ages that were found, or a random selection?

But, even so, we're still measuring words that occur not possible words. Any language with a relatively complex syllable structure will have many gaps - syllables that are phonologically acceptable but not in use. We know Voynich words have little structural variety. Could your tests be run on Pinyin?

---

(04-02-2016, 10:56 AM)-Job- Wrote: You are not allowed to view links. Register or Login to view....

Given that the glyphs for "m" and "r" are so similar, it's possible that some of these pairs represent the same character.

In any case, if they are in fact different characters, we can at least rule out some vowel/consonant assignments.

...

What are your thoughts on the issue in terms of sound assignments? Were [r] the sound /n/, [m] could conceivably be /ng/. This would explain 1) its existence almost exclusively at the end of words, 2) its apparent variability with /n/, and 3) the similarity of the characters (assuming that the script was devised with the goal of showing such relationships). Naturally, this would force [l] to be /m/.
It's something I've thought about, but have no good evidence to back up. So it is just a guess for now.

(04-02-2016, 08:09 PM)Emma May Smith Wrote: You are not allowed to view links. Register or Login to view.Job, your statistics are very interesting. For the letter replacements in other languages, were these the highest %ages that were found, or a random selection?

The highest values (e.g. above 5%).

(04-02-2016, 08:09 PM)Emma May Smith Wrote: You are not allowed to view links. Register or Login to view.But, even so, we're still measuring words that occur not possible words. Any language with a relatively complex syllable structure will have many gaps - syllables that are phonologically acceptable but not in use. We know Voynich words have little structural variety. Could your tests be run on Pinyin?

For Pinyin, using You are not allowed to view links. Register or Login to view.:

k replaced by g: 8/109 7.0%
m replaced by b: 11/166 7.0%
t replaced by d: 14/194 7.0%
ā replaced by à: 29/363 8.0%
ē replaced by à: 10/142 7.0%
ō replaced by ò: 9/105 9.0%
ǎ replaced by à: 25/326 8.0%

The character set was the following: abcdefghijklmnopqrstuwxyzàáèéìíòóùúāēěīōūǎǐǒǔǘǚǜ

Well then, that's very interesting. I'm astounded that the values for Pinyin are the same, or even lower, than for English.

For pinyin it's going to matter a lot whether you're looking at multi-syllabic words, or monosyllables.  The above results must be for words.  If you look only at monosyllables the result is going to be very high, and if you ignore tone it will be 100% in many cases.

Job's replacement lists here do seem to cut to the heart of something very specific to Voynichese: its curious substitutibility, which makes almost no sense if you stubbornly try to view Voynichese as a natural language.

To my eyes, though, these are merely substitutions within the Voynichese's cover text, where many of the plaintext letters and numbers are enciphered as verbose groups, e.g. or / ol / ar / al.

Hence, you would need to pre-parse Voynichese into verbose groups before you would start to get more genuinely revealing substitutibility tables: and this is more generally true, that all the while you persist in seeing (for example) 'o' and 'a' as single letters, Voynichese can (and will) make no sense to you.


But I suspect that even this doesn't quite explain the specific substitutibility of t and k. A decade ago, I proposed that these might instead be 'substitution tokens', markers for copying one or more letters from elsewhere on the page (or possibly even from a different page). But that's another story entirely... :-)

(12-02-2016, 05:50 PM)nickpelling Wrote: You are not allowed to view links. Register or Login to view.To my eyes, though, these are merely substitutions within the Voynichese's cover text, where many of the plaintext letters and numbers are enciphered as verbose groups, e.g. or / ol / ar / al.

In my opinion there is a more generic process at work, which results in a large variety of similar words, from which the abnormal character replaceability values emerge.

There's definitely some further work to be done here. For example, some characters can be removed entirely, without producing an invalid word.

I have not done a comparative analysis of character removals yet, but i have You are not allowed to view links. Register or Login to view. which illustrate what happens when a character is removed.

In those overlays, words highlighted in white become invalid when the target character is removed, while words in blue remain valid.

For example, we can see that EVA 'n' is not removable:
You are not allowed to view links. Register or Login to view.

And character 'a' is far less removable than 'e':
a: You are not allowed to view links. Register or Login to view.
e: You are not allowed to view links. Register or Login to view.

In fact 'e' seems suspicious in this regard - it's almost optional - though this is possibly because 'ee' is replaceable by 'e'.

I will post some actual numbers for Voynichese and other sample texts later on (maybe in a separate thread).

On a different topic, i noticed a familiar tail stroke in the 'p' characters of You are not allowed to view links. Register or Login to view., and thought it was worth mentioning:
[attachment=108]

---

BTW, just to drive home the message about word variability.

In this overlay, words in blue have at least one variant in the text that differs by a single character:
You are not allowed to view links. Register or Login to view.

Without saying that this is the solution, but just to give an example and some food for thought:

if the words were all Roman numbers (i ... ix ... lvi ... cliii) 
the same feature would emerge. Each word would have several ones that differ from it by
only one addition, deletion, replacement or swap.