Vord length distribution

[MarcoP]

What about syllabification? I think syllables in some languages may have a distribution similar to Voynichese.

That's what Stolfi seems to build his argumentation upon, but as I said I could not fully understand his idea. When it comes to linguistics, especially beyond Russian and English, I do not feel very comfortable.

Hi Anton,
I don't know much about linguistics either, but here is how I understand what Stolfi wrote. Of course, I might have misunderstood something.
I managed to come rather close to Stolfi's results with the Vietnamese VIQR writing system. The Viet text he used is online You are not allowed to view links. Register or Login to view..
I converted it to UTF8 and removed punctuation.
This graph shows the distribution together with Voynichese in the Currier-D'Imperio transliteration. As one can see, the two curves are quite similar, with Viet words just being a little shorter. 

   

Words in the Viet file look like this (random sample):
tu'ng
do^'i
dde^`
ky`
lo`i
cheßo

The Latin characters correspond to vowels and consonants (sometimes two Latin characters form a digraph corresponding  to a single sound). All the other characters (Greek letters included) are modifiers that apply to the preceding Latin character. The Greek letters where introduced by Stolfi and replace other VIQR symbols that normally represent punctuation.
In Vietnamese, each word is a syllable. A syllable must contain a group of vowels that can be preceded and followed by groups of consonants, forming the well-known CVC structure. This happens in most (all?) languages and can also be observed in Vietnamese. The first word in the list above (tu'ng) includes a Consonant (t) a Vowel (u') and two more Consonants (ng).
Each of the three components CVC has a variable length, with only the Vowel component constrained to be non-empty. These three components vary independently. In Viet, you can have words like "i"(a single V) or "thu+o+`ng" (2C, 2V with modifiers, 2C). Since the structure of a syllable is basically limited by human physiology, it cannot be arbitrarily long: the maximum length (which depends on both the language and the writing system) could correspond to N in your success/failure example. 

---

Rene's last post and the idea that combining binomial distributions can result in a new binomial distribution suggested to me that also dictionaries with a larger but still limited number of syllables might be worth considering. I experimented with Shakespeare's English. I considered a,e,i,o,u,y as Vowels and all other characters as Consonants.
I created three files for 1 syllable (CVC), up to 2 syllables (CVCVC) and up to 3 syllables (CVCVCVC).
These are the longest words in each file:
1 strengths
2 heartstrings
3 transgressions

As can be seen from the following graph, the three files show symmetrical distributions.

   

The curve for 1 syllable English words (orange) is very similar to that for Viet (but with shorter words, since English does not include vowel modifiers).
I did not expect the curve for 3 syllables (green) to be such a good match for EVA-encoded Voynichese (ZL transliteration, no uncertain spaces). 

The 3-syllables plot is not too different from the curve for the whole Shakespeare text, but the difference seems to be exactly what is needed to make the curve symmetrical. The difference is of course due to the fact that the original text contains words up to six syllables long (e.g. 'impossibilities' 'unnecessarily').
   
So, unless I made some errors in the process, it seems that limiting the number of syllables can turn the distribution of a dictionary into a symmetrical shape. We have already seen that scribal abbreviations do not seem to work (the Bonavenutra text discussed above) and, as Jonas wrote, a simple threshold on word length also wouldn't work:

If you constrain the length you will only cut off the tail

One of the reasons why I find this result interesting is that it fits with You are not allowed to view links. Register or Login to view..  According to her analysis, Voynichese words are constrained to include at most three syllables. I was not aware that such a constraint could result in a binomial distribution of dictionary word lengths.
Of course Stolfi's "Chinese" (or more generally "monosyllabic") hypothesis has the advantage of pointing out actual linguistic texts with symmetrical distributions. I don't know if natural languages which limit words to two or three syllables exist. But Emma's approach has the advantage of being based on the actual structure of Voynichese words.