Let me restate and expand a point I made on my VMD talk, about "agnostic research".

It has been stated that research about the VMS, such as statistical or structural analysis of the text, should avoid making any prior assumptions about the origin and nature of the manuscrpt.  Any such theories, not based on any data, would be pointless, nedlessly polemical, "unscientific" hence damaging to the field's reputation, and so on.  They may create confirmation bias, where one looks only for evidence that confirms the assumptions, ignoring evidence that contradicts it.  We should instead wait for the manuscript to be deciphered; and then the text will either tell about the origin, or give enough clues for us to deduce it.

Those are quite valid concerns.  Unfortunately, "agnostic" research is not a viable option.  One needs to make at least some assumptions about origin and nature in order to choose which statistics to collect and which structures to look for.  Without that guidance, one is likely to end up with a big confusing pile of numbers and diagrams from which one cannot draw any clear conclusion.  And that unfortunately has been the fate of much VMS research to date.

And even then one cannot avoid the problems listed above.  For instance, I am afraid that those who really believe in Currier's "two languages" have succumbed to confirmation bias -- by concentrating on the differences between the two word distributions, while ignoring the much more numerous similarities.  Because I think that early on their intuition told them (consciously or unconsciously) that there were two authors and/or two languages.

To avoid that fate, one should instead try the old and tired "scientific method": make some conjecture about the issue, then devise and execute the most effective (cheap and categorical) experiment that could confirm or refute [i]that conjecture[/i].  The hypothesis need not include a specific language or encryption scheme, but it must be more specific than "it could be anything".

I would say that the few firm results of the last 60 years of research on the text were obtained in this way.  Effectively, very early on the conjecture was formulated that the text could be an European language, written in its original spelling system but with an invented alphabet --- that is, in a simple substitution cipher.  This conjecture was quickly disproved by several simple tests, including 

----------------------------------------------------------------------

I believe that any statistical peculiarities of the head lines can be explained by (a) the parag text having a fixed formula; and/or (b) the "puffing" transformation which was applied by the Scribe to the head line, without regard for its contents, just because of his scribal habits.

Cause (a) can result in the word distribution at the begining of the paragraph being arbitrarily different from that of the body, or of running text in general.  And, since glyph and diglyph frequencies are dominated by their occurrences in the most fequent words, these too can be modified by cause (a).

The "puffing" transformation (b) may be not quite trivial, not deterministic, and not invertible.  Like the capitalization of headlines and titles in modern English publications.

You should have seen the recent posts on this forum where people report running the simplest and dumbest line-breaking algorithm on various texts, including English, and find that it causes the first word of each line to be longer than average, while the last few words of each line to be shorter than average.  And, again, since the frequency or glyphs and diglyphs is dominated by their occurrence in the most frequent words, and these tend to be shorter than average, that same cause can significantly change those frequencies too.

And (in manuscripts, but not in printed text) there is also the tendency of words to be squeezed together or stretched out when the writing gets near the end of a line.  This effect may cause some otherwise rare words and diglyph pairs to appear when transcribers read twowords asasingleone or an ov erly stretc hed word as two separate words.

Stepping back, it should be clear that (A) all sorts of anomalies are expected to exist on head lines and at the start and end of body lines, and (B) those anomalies may be produced by many different causes, many or all of them being irrelevant to the goal of deciphering the manuscript.

[quote][quote]That would be ideal, but (as I pointed out in my VMD talk) such an "agnostic" approach is unfeasible.[/quote]You and me are never going to agree on this in a million million years.  ;)  I can study line patterns without having a theory or belief that Voynichese is a particular language or in a language family.[/quote]

I hope we can agree well before that. :)

You surely can do agnostic analysis, but it is very likely that it will be a frustrating waste of time, producing a big pile of data that you cannot make heads or tails of.  I seriously insist that you follow the "sceintific method" -- make a conjecture, then devise experiments that will most effectively (cheaply and categorically) test it.

The conjecture need not imply a specific language or language family, but it must be more than "could be anything".

Consider the "line-start" and "line-end effects" above.  For 60+ years people have been "agnostically" collecting statistics of words and glyphs at the ends of lines, and got nowhere.  But once someone conjectured that the cause for those discrepancies could be the line-breaking algorithm, it took minutes to run the experiments that confirmed the conjecture.

So my advice for further research on this line would be to pursue two more specific goals:

  (1) figure out the Scribe's "puffing algorithm" so that we can un-puff the head lines
  (even it not deterministically) for the benefit of further studies.
 
  (2) determine whether the head lines have different word distributions (apart from the "puffing") because of "fixed formula" effects - like the parag starting with a list of names for the plant, and/or other words of a specific class, hence with no function words, verbs, etc..
 
A possible test for (2) would be to take pages where the headline of the first parag is much shorter than normal, because of narrow margins or intruding figures, collect the distribution of the first word(s) of the second line, and compare it to that of the first word(s) of lines 4 onwards.  The justification being that, on those parags, any "fixed formula" effect is likely to spill over into the second line.

Another possible test would be to consider pages which have only one plant but two parags with full-width head lines, anc compare the word distributions of the two head lines.  The justification being that any "formula effect" is likely to affect the first parag head line but not the second one.

Can you think of better tests?
 
As for (1), my proposed conjecture is that "puffing" consisted of blindly replacing certain glyph combinations, probably but not surely containing tike gallows, by other glyph combinations with puffs gallows.  Note that the replacements may have been incomplete and inconsistent.  My idea to test this conjecture would be to compare the word distribution within parag head lines (excluding the first word, which seems to be a hapax) with the word distribution within parag bodies, and try to match at last some of the words of the former, with puffs, with words of the latter, without puffs.  Perhaps generate several plausible replacement rules and see which one can maximize the agreement between the two word distributions.  But, again, perhaps you can figure out a better way to test this theory.

All the best, --jorge