Well, Kevin Knight gave his Voynich lecture (it’s the one I mentioned a few days ago), and an attendee (Jeffrey Shallit) has been kind enough to post a high-speed précis of what Kevin Knight said onto his “Recursivity” blog. Of course, since you already know the basics, I can strip that down even further. 🙂
As I mentioned before, Kevin Knight’s approach is based around using expectation maximization to try to algorithmically categorize the letters in the alphabet into one of two clusters – basically, vowels and consonants. However, despite the looming vowel-like presence of ‘a’, ‘e’, ‘i’, and ‘o’ (that so convinces linguists), this approach fails to produce “anything particularly meaningful” for Voynichese.
His more recent research has tried the same trick but with words instead of letters. According to Jeffrey Shallitt’s post:-
When you do so, you get two clusters: the words in the “herbal”, “astrological”, and “pharmacological” sections predominantly fall into one cluster, and the words in the “biological” and “cosmological” sections predominantly fall into another.
However, Knight’s attempts to further subdivide these two clusters failed to produce any (linguistically) helpful clusters.
I think that while this is manifestly unhelpful to linguistic VMs theories, it may yet prove to be codicologically very productive. My position is that Voynichese evolved over time, and that the VMs itself was composed in a number of writing phases: it may well be that the clusters that are emerging are clusters in time, not in content. Could it be that this kind of cluster analysis could be used to reconstruct the developmental arc of Voynichese?
I hope that Kevin Knight decides to make his clustering results available, so that they may be assessed in the context of quite different (non-linguistic) ways of looking at it…