Definitely there are places where it would be of value to have more visual information about. Let's say the writings in You are not allowed to view links.
Register or
Login to view. or many places in the MS where (it seems that) something has been erased.
Naked eye is better than any scan in this respect, because scan is two-dimensional only, while when you're looking at the object with your own eyes, you can look at different angles and see much more.
(07-07-2021, 08:00 PM)Anton Wrote: You are not allowed to view links. Register or Login to view.Definitely there are places where it would be of value to have more visual information about. Let's say the writings in You are not allowed to view links. Register or Login to view. or many places in the MS where (it seems that) something has been erased.
Naked eye is better than any scan in this respect, because scan is two-dimensional only, while when you're looking at the object with your own eyes, you can look at different angles and see much more.
I think there is an understandable tendency to see criticism of the resolution of the currently available scans as a sign of ingratitude. The quality of the current scans is so much greater than there was before when only crude black-white scans seemed to exist. In fact I was very impressed when I first saw the quality of scans of the Voynich.
However the reality is as you say that there are definitely places where it would be a value to have more visual information; in fact I would suspect there are probably very many places where in it would be of value.
Maybe if people haven't made progress in the coming decades the manuscript will be rescanned/photographed in 3D and at a higher resolution. The scanning technology applied to the antikythera mechanism shows the kind of thing that could potentially be done and those scans have been crucial to the understanding of that object.
Deciding what each of the following is exemplifies the problem.
[
attachment=5639] - Must be an EVA-o
[
attachment=5638] - Is this an EVA-a ?
[
attachment=5637] -
Is this an EVA-a ?
[
attachment=5636] - Must be an EVA-a
These are taken from the Rosettes T-O Map
When you look at things at this resolution the quality of the scans seems terrible.
If a computer were given all images for each instance of either EVA-a or EVA-o (or EVA-c?) in the manuscript and then tasked with splitting them into 2 sets the EVA-a ones and EVA-o ones it might fair quite well.
I was struck by the standard of image recognition software now available when looking at websites to identify plants and this kind of thing does not seem unrealistic, though implementing software to transcribe the whole text sounds like a lot of work, hopefully in the future there will be more tools to do that.
I have a strong suspicion that a computer could probably do a much better job than a human of transcribing the Voynich as it seems likely that the human transcriptions are not very good. Obviously the quality of transcriptions is vital to Voynich research.
(09-07-2021, 05:31 PM)Mark Knowles Wrote: You are not allowed to view links. Register or Login to view.I have a strong suspicion that a computer could probably do a much better job than a human of transcribing the Voynich as it seems likely that the human transcriptions are not very good. Obviously the quality of transcriptions is vital to Voynich research.
I actually think that the quality of transcription is not very important for VMS research. Even the most ideal transcription won't affect key properties such as word/letter frequencies greatly.
A good theory cannot depend entirely on the exact transcription of 1, 3 or 23 words - they could be errors, proper names, foreign words. Either make assumptions about these words or take another piece of text (there is a lot of text out there) - in any case it is the next step that really matters.
(09-07-2021, 06:07 PM)farmerjohn Wrote: You are not allowed to view links. Register or Login to view.I actually think that the quality of transcription is not very important for VMS research.
This is surprisingly often true. Since the VM text is so large, you can get away with a pretty large amount of errors. Or put differently, while there are many differences between transcriptions, the impact on statistics is often much smaller than one would intuitively think.
Now on the other hand it's surely annoying if one's favorite label happens to be unclear.
(09-07-2021, 06:07 PM)farmerjohn Wrote: You are not allowed to view links. Register or Login to view.I actually think that the quality of transcription is not very important for VMS research. Even the most ideal transcription won't affect key properties such as word/letter frequencies greatly.
Well, there could be a tendency throughout the manuscript by transcribers to misidentify an EVA-a as an EVA-o And therefore an improved transcription could make a difference to the frequencies of EVA-a In fact when it comes to vords only 1 character needs to be misidentified for the whole word to be misidentified.
(09-07-2021, 06:07 PM)farmerjohn Wrote: You are not allowed to view links. Register or Login to view.
A good theory cannot depend entirely on the exact transcription of 1, 3 or 23 words - they could be errors, proper names, foreign words. Either make assumptions about these words or take another piece of text (there is a lot of text out there) - in any case it is the next step that really matters.
A standard approach to deciphering a text is to construct a crib, this could be 1, 3 or 23 words. The fewer the errors the better the crib. I have found fewer than 10 words that I know that I consider could possibly be used as crib words. The history of Voynich research and the construction of possible cribs is not a good one, but any crib word could help. Although I am inclined, given my perspective on the nature of the cipher, to the view that quite a lot of crib words would be required in and of themselves to decipher the Voynich. However a crib could be coupled with other techniques.
I am of the opinion that the following, from the rosettes page, shows the directions North, South, East and West in abbreviated form. This would seem to contradict other discussed directions on this page, but I see these as local directions not whole page directions. (I think all the characters are clear except the 3rd character of the top word.)
[
attachment=5642]
(09-07-2021, 06:24 PM)Koen G Wrote: You are not allowed to view links. Register or Login to view. (09-07-2021, 06:07 PM)farmerjohn Wrote: You are not allowed to view links. Register or Login to view.I actually think that the quality of transcription is not very important for VMS research.
This is surprisingly often true. Since the VM text is so large, you can get away with a pretty large amount of errors. Or put differently, while there are many differences between transcriptions, the impact on statistics is often much smaller than one would intuitively think.
What you say makes sense given the law of large numbers, errors are evened out. However that is the case if the errors are not often in a particular consistent direction, which they may be.
(09-07-2021, 06:24 PM)Koen G Wrote: You are not allowed to view links. Register or Login to view.
Now on the other hand it's surely annoying if one's favorite label happens to be unclear.
Yes, this is certainly particularly a concern of mine. If one is interested in very specific pieces text this can become a problem.
(09-07-2021, 05:31 PM)Mark Knowles Wrote: You are not allowed to view links. Register or Login to view.I have a strong suspicion that a computer could probably do a much better job than a human of transcribing the Voynich as it seems likely that the human transcriptions are not very good. Obviously the quality of transcriptions is vital to Voynich research.
As I understand it, the main problems with transcriptions are ambiguity and lack of consensus on modified characters and spacing. Machine learning shouldn't help with these per se.
But, a machine learning transcription of the ms is overdue for a few reasons. Thinking about it for a second, it would also make easy probability ranges for everything on the page, instead of what all current transcriptions do, which is bite the bullet and put down a 1:1 transcription for what you think are the characters, even though it's really not always clear. Surely this type of transcription has been thought of before, so why isn't there one?
In page 1r, 2nd paragraph, under the red drawing on the left, ther is a message written in very small graphs, the superimposed diacritical mark looks like a heart but if you pay more attention its a composition of letters.
The resolution is to low to read all.
Do you think that if we let the Beinecke Library know, they will make a higher resolution scan?