Automated Topic Analysis of the Voynich Manuscript

[MarcoP]

Hi quimqu, as I pointed out You are not allowed to view links. Register or Login to view., this appears to overlap with the ongoing research by Lisa Fagin Davis and Colin Layfield. We know that quires were put together in an at least partly arbitrary way; trying to understand more of the order in which the ms was created is extremely interesting.

Most of the first 1/3 of the herbal, it is almost pure A. But then gruadually, paragraphs that mix B topic words appear, and appear quite strongly. They are not as continuous as strong topic A paragraphs, but they appear suddenly after 1/3 of herbal paragraphs. Are the herbs starting to be described partially with topic B?

We know what is happening here since Currier’s analysis half a century ago: herbal bifolios by different scribes were mixed and bound together. This is detailed in You are not allowed to view links. Register or Login to view. (Table 1). All pages from f1 to f25 (“the first 1/3 of the herbal”) were created by scribe 1. After that point, scribe 2 pages begin to be inter-mixed with scribe 1 pages. If we take the illustrations as indicative of a topic, the change in statistics does not appear to be due to a different topic, but to a different scribe.

The CUVA bigram plots You are not allowed to view links. Register or Login to view. (or before) are a good match for your topic line. See bottom of the page.
   
E.g. the plot for ‘ed’ shows how the zodiac section (gray) appears to gradually shift from Currier A (bottom) towards Currier B (top). It also shows that Quire13 Bio is more strongly “B” than Quire20 Star-Paragraphs. It also shows that Pharma (yellow) is comparable with Herbal A pages (both HA and Pharma are attributed to Scribe 1).
In general, the fact that there are such huge differences at bigram level does not sound like different topics, but rather like different languages (like Portuguese vs Spanish, as you mention). But of course the gradual shift between the two A-B extremes does not fit with the idea of different languages (and we know that Voynich bigrams cannot correspond to bigrams in European languages).

   
As Rene’s plots show, this information can be used to reorder Voynich sections, so that the text begins with Strong-A (HA) and ends with Strong-B (Bio). If I understand correctly, Lisa and Colin are going even more in-depth, also considering the stains that affect many of the pages; they are working on bifolio-level reordering, rather than section-level. I expect that their paper will be a major step forward in our understanding of the structure of the text.

EDIT: another point that I think Rene mentioned in the past. Results based on only a few samples are more noisy and unreliable than results based on larger sets. This could play a role in the fact that Bio/Q13 paragraphs get more consistent results than the much shorter Stars/Q20 paragraphs.