[Blog Post] Rightward and Downward in the Voynich Manuscript - Patrick Feaster

[pfeaster]

I was wondering if single-scribe corpora are large enough for detailed word-oriented analysis. For instance (unless I am miscounting), each of Scribe2 and Scribe3 wrote about 150 occurrences of daiin. This means that, when considering recto and verso separately, each cell in a 10x10 grid averagely represents less than one occurrence on average.

You make a good point, and while I'm not sure what tests of statistical significance would be most useful, I can supply one more piece of data.  In these displays, the numerical value of pixel brightness is proportional in a linear way to the number of tokens at each point.  Tokens in shorter lines and shorter paragraphs have a bigger footprint, but when I've tried to adjust for that by scaling brightness in inverse proportion to footprint size, it hasn't seemed to make much of a difference, so I stopped doing it.  So the brightest areas are now simply those where the greatest number of token footprints overlap.  The numerical value of each point before I boost contrast should approximately equal the number of tokens contributing to it, although because of the spline interpolation, a "direct hit" on the center of a token's position receives a higher score than an oblique one, so this won't be exact.

For the four charts of [daiin], the maximum values were originally:

Scribe 2 recto: 5.3
Scribe 2 verso: 3.1
Scribe 3 recto: 8.1
Scribe 3 verso: 16.0

So in the Scribe 3 verso image, the brightness scale runs from 16 tokens at brightest to 0 tokens at darkest, while in the Scribe 2 verso image, it runs from only around 3 tokens at brightest to 0 tokens at darkest (which I'll admit isn't much of a range).

For the qo+gallows / o+gallows charts, the maximum values were:

Scribe 2 recto: 14.2
Scribe 2 verso: 15.0
Scribe 3 recto: 34.0
Scribe 3 verso: 32.0

I'm also subtracting the minimum value before rescaling, but I'm pretty sure that in each of these cases there's at least one point with no tokens, so that the "real" minimum is indeed zero.