The Voynich Ninja

I was recently thinking about Palindromes (words that read the same backwards and forwards).
All European languages (that I know of) have single word palindromes, but this effect seems to be almost absent in the VM.
You are not allowed to view links. Register or Login to view. are simple three letter words.
The only longer palindromes seem to be unique in all cases with the exception of occo, which appears three times in the manuscript.
Here are the ones I've spotted (quite possible I've missed some, this was only a quick count)
dydyd (f1)
seees (f3)
oeeo (f6v, f72v2, f101v2)
ykaky (f55v)
yekey (f69v)
ylaly (f73v)
lolol (f72v)
shchs (f113r)

The low number of palindromes is of course to be expected, due to the position awareness of glyphs.

It's possible that such palindromes are actually the result of misspellings, and this could give us some concrete examples of such scribal errors within the corpus, allowing us to correct scribal errors and reduce erroneous words from the transcription.

For example, taken at random because it made me laugh out loud, lolol: lo appears 15 times by itself, 182 as a word initial and ol 3052 times as a word final and 538 times as a word by itself. But lo*ol only gives two results, lolol and lolkeol. This second word is more likely to be two words run together, as both lol and keol are common words. This suggests to me that lo and ol have well defined functions, but shouldn't be used together; the scribe made a mistake with lolol, and missed out a space in lolkeol. What mistake in lolol? Well, the prior word is checkho, which is unique. If we move the first l over, then we get checkhol, which appears twice in the corpus. We now have checkhol appearing three times, followed by olol, which appears 18 times in the corpus.

So we have now removed three unique words from the corpus in a logical manner!

No idea if we can do this with the rest of them, it's getting late and I'm tired now. Has anyone any research into this angle, or into reducing the number of unique words by seeing if they can be exploded and reassembled with adjoining words?

As Grove pointed out. Not quite the same as this, as here we don't have to drop any glyphs. 

I've been pondering for some time about creating a spell checked transcript, one in which we fix as many uniques as can be easily transformed into valid tokens.

You know, it never occurred to me to scrub the text. I go almost the opposite way, looking for anomalies to see if they are clues.

Let's say you have a common word in English, like "bend". It might show up hundreds of times. Then there's a similar word, but less commonly used, like "wend". It can be uncommon but still perfectly valid. Maybe it's because of patterns like this that it never occurred to me to scrub the text.

There might be reasons to do a sanitized version of the transcript, however. It's something I need to think about for a while. As I mentioned, my perception of the text tends to be in the other direction. I tend to try to retain anomalies or rarities, as long as they are clearly written.

A scrubbed version can be handy if you're looking for multi-word patterns for example. That is, if you think the cleaning can be done in an acceptable way.

[attachment=3780]
Based on what I have already written and what follows a possible system.
self control by unum cus and unum cum. cum have a lot of.

Example without guarantee

[attachment=3781]
Example: