[Article] Glyph combinations across word breaks in the Voynich Manuscript

[MarcoP]

Emma, a couple of quick and probably unimportant questions upon the transcription preparations:
I am assuming that labels were stripped out of the transcriptions?
What about the "paragraph initials" standalone glyphs, such as appear on ie, f66r?

Hi David,
thank you for your comments.
Most of the analysis was performed on Takahashi's transcription. All word-pairs that appear consecutively on the same line of the transcription file were considered. Labels appear in individual lines, so they were mostly ignored, with the exception of those that include more than one word.
The glyphs in You are not allowed to view links. Register or Login to view. are also transcribed separately, one per line, so they were ignored too.

I was wondering what results you would get if you repeated the experiment against the labels, putting them into their natural order and assuming the "space" exists between them.
One possible angle to check: Assuming the labels were written down in sequence, one would assume that under the auto-copying hypothesis, you would see similar stats to the main text, as the scribe would use the same procedure to generate them. 

Alternatively, if the labels have different stats than the main text, then it suggests that there are two separate text generation methods at work.

As discussed You are not allowed to view links. Register or Login to view., simple repetition cannot be the reason for the "biased" word-break combinations in the VMS. Also, the fact that q- is almost absent in labels suggests that labels and paragraph-text are not identical (even if they share "morphology" and almost half of the labels appear in paragraph-text, as discussed You are not allowed to view links. Register or Login to view.). 
Edit: see also You are not allowed to view links. Register or Login to view..

I have tried to follow your suggestion. I concatenated all labels, generating a single line for each page. I used the Zandbergen-Landini transcription, excluding all paragraphs and circular / radial text. Uncertain spaces were ignored.
These are the results for the ten most expected combinations, with the corresponding actual counts.

   

This histogram shows the deviations from the expected values and can be compared with Chart 1 in the pre-print. Obviously, we are speaking of small numbers: deviations might not always be significant.
The high values for -y.o- could be due to the You are not allowed to view links. Register or Login to view.. Anyway, the deviation from the expected count is not large (113%).