- Options
- Login
- Desktop Style
Options
Elephant in the Room Solution Considerations
- Elephant in the Room Solution Considerations
-
RE: Elephant in the Room Solution Considerations
MHTamdgidi_(Behrooz) > Today, 01:57 AM
@Jorge_Stolfi, thanks for your considerations.
Let’s begin from your last paragraph, when we do know the language. We can see ‘The’ and ‘the’ as being the same word because we know the language. Otherwise, we do not have that privilege, which is the case in the VM. So, we must treat the transliteration in a way that would distinguish between The and the, unfortunately. There is no other way to do it, in my view.
T, h, and e are separate letters in both versions, because we see that in our supposed original. If they had been connected (not knowing what they mean), we have no choice but respect the original and treat them as connected. If c-c or a c-^-c (a c-c with a diacritic on top) is all one connected symbol, we need to treat it as that and respect the original, rather than assuming it is comprised of the three parts. Of course it is helpful to see if a k between a double c is used separately on its own as well, and v101 does that too (which is why I like it more than others). We will surely note that in that one symbol, the author/scribe is using a c or a gallow to construct it; but still, there is a reason they are visually rendered as one symbol. So, yes, also if we see different looking diacritics for the c-c, even when we are not sure, we should take that into account, unfortunately, since we are dealing with an unknown language.
It is only when we respect that visual, that we can then compare it to a c-c and say, “it seems this one is the same as the other with this or that diacritic on top.” The same holds for “bench” symbols. By splitting it into the three parts, we have already allowed our own assumptions dictate the visual original’s features. We don’t know if the author and the scribe were agreed on that, but that is something we cannot consider as an assumption to prefer this or that transliteration system.
For this reason, following the visuals of the original are essential. Here we arrive at your listed assumptions and conditions.
I do understand your points regarding the assumption variations. But the problem is that they are all assumptions, and we do not know whether any is truer than the others, even though you and I may prefer some over others. To be empirically strict, yes, we should treat noticeably different symbols (even if we suspect it is a different handwriting style) as being different, though see them as possible variations of the same symbol.
Yes, there are challenges the VM poses: 1-The original can be illegible; 2-The retracing may be in error; 3-Author intentions may have been distorted in the scribing; 4-handwriting styles may add undeterminable noise for transliterations choices, and so on.
If it is hard to decide whether a letter is one of two similarly looking ones, that can also affect the word in which it is found. So, the ambiguity would exist no matter what transliteration systems we devise.
But all of the above have less to do with the main point I am trying to convey.
The final goal is to be able to read the VM text, not any transliterations of it, no matter how good or bad the transliteration systems are.
Yes, their helpfulness depends on whether they can most immediately convey the visual characteristics of the original. A double c is a double c, connected on top, not two separate letters. The author and/or scribe wanted to say c-c as one symbol, since it was possibly a contraction for another word or set of words. A bench is all one symbol, not split in three parts as assumed. V101 at least acknowledges the double-c with diacritic as being one symbol, transliterated as 2.
The MAIN problem is when any of these transliteration systems are used to draw linguistic conclusions about the language, i.e., whether it is a natural, constructed, hoax, gibberish, etc., language. You can’t make such judgments based on transliterations, no matter how good or bad they are, whether v101 or EVA.
I have no problem with transliteration systems being used to know statistically how many of this or that symbol appears, alone or in combination in the VM, assuming all the imperfections involved. This of course assumes we can read the text (legibility), scribal handwriting, scribal or other retracing, how faithful or aware the scribes were about a (living or dead) author’s intentions. These complicate constructing such transliteration systems.
But ultimately the criterion should be the study of the actual original as visually displayed in only manuscript copy, we have.
I think preoccupation with transliterations have prevented us from paying more attention to the actual visual information the original text is offering. The more we spend time on the former, the less we spend time on the latter.
The problem I see (and this can address @Renez as well) is that analyses of transliterations, good or bad, for making linguistic judgments about the original. That is where transliteration systems can prevent the study of the text itself. You -
RE: Elephant in the Room Solution Considerations
Jorge_Stolfi > 11 hours ago
(Today, 01:57 AM)MHTamdgidi_(Behrooz) Wrote: You are not allowed to view links. Register or Login to view.following the visuals of the original are essential.
There are two distinct "flaws" of EVA that you and others seem bothered about:
(1) It uses more than one letter to record a single glyph (a "maximal set of connected pen strokes")
(2) It does not record certain details of the glyphs, like the shape and position of the plume on the Sh.
I addressed (2) in the previous message.
Yes, whether (2) is good or bad depends on the assumptions one makes about the way the book was written and the intentions of the Author. As I said in my Voynich Day talk, it would be nice if we could study the VMS "agnostically" -- without making any such assumptions beforehand; but that is impossible. One must make prior assumptions in order to do any research; in any field, not just on the VMS. One cannot get out of bed in the morning without making the assumption that the floor is still there where it was the night before, and did not turn into a big pool of ketchup.
Even if you use VT101, you are making many such assumptions, like "the size of the glyphs does not matter, only their shape", "there are only two kinds of space, inter-glyph and inter-word", "the spacing between the lines does not matter", "it does not matter if a line ends a few mm before or after the right rail", etc. And "the plume on a bench glyph can have only one of these six shapes".
As for (1): indeed, using two or more Latin letters to encode what we assume is a single "letter" of the Voynichese script is an annoyance for some types of computer processing, like computing the frequency of "letters".
But almost every alphabetic script in the world uses two or more letters of its alphabet to denote certain single sounds: "sh" in English, "gn" in Italian, "ch" and "oe" in German, "ll" in Welsh, ... People who are native speakers of those languages are used to that kind of encoding trick since kindergarten, and thus are hardly bothered by EVA using it too.
EVA does record whether glyphs are connected or not, through certain simple rules. Thus "ch"or "Ch" is always a single connected glyph (Ch), and so is "CTh" (CTh); whereas "ee" is always two separate glyphs (ee), and "ete" is three (ete). Besides, Ch is two separate pen strokes, and CTh is four, which makes their EVA encoding seem almost logical.
On the other hand, that feature of EVA makes it possible to write all the common glyphs using only the 26 lowercase Latin letters; which is a big plus for computer processing. Moreover, it makes it possible to record many rare glyphs, like CTy ("CTy"), Cky ("Cky") or CTHh ("CTHh") without using additional computer codes or rules. Additional codes are needed only for rare glyphs, that occur only once or twice in the whole book; and those could all be encoded as "?" without significant harm for most analyses.
And, finally, the EVA design decision to use only Latin letters made it possible to "pronounce" the words. This genial idea of Jacques Guy is very helpful when transcribing. One can mentally "read" a whole word from the scanned image as a string of sounds, and then type that into a file, as one would do when transcribing any other alphabetical language. That is normally much faster and reliable than trying to memorize the shapes of the glyphs. That is why the people who study Sumerian or Ancient Egyptian have made up sounds for all the glyph of those scripts, even though in many cases we have no idea of what those glyphs sounded like (or even if they were "letters" at all).
And, of course, everybody who has been in this corner of the internet for more than a week knows that we have absolute no idea of how the Author pronounced daiin, but it surely was not even remotely close to "die in".
In conclusion, which encoding is best for you depends on the assumptions you make and what sort of analysis you want to make. Fortunately Rene's tools let you choose between any of the historical or current encodings; and they could easily accommodate more. "The great thing about standards is that there are so many to choose from"...
All the best, --stolfi -
RE: Elephant in the Room Solution Considerations
ReneZ > 10 hours ago
-
RE: Elephant in the Room Solution Considerations
Antonio García Jiménez > 7 hours ago
I remember Gabriel Landini joining the forum during a discussion I reopened about Zipf's law. It was last summer, just shortly before his death. I'm sorry for your loss, René. He must have been a good friend of yours. -
RE: Elephant in the Room Solution Considerations
MHTamdgidi_(Behrooz) > 10 minutes ago
@ ReneZ, Antonio García Jiménez, Jorge_Stolfi, and colleagues, please accept my condolences for losing a good friend in Gabriel Landini. I was also sad to learn (at the beginning of my research in the VM last year) that Stephen Bax also passed away in 2017.
@ Jorge_Stolfi, thanks again for your rejoinder. I understand your point overall is there are many options to choose from among transliteration systems depending on what we wish to do in research and assumptions we have (ones that in your opinion are inevitable) or preferences we may have for having a “pronunciation ability” value in one or another system.
Of course all can do what they wish, but we may also have differing views as to which system can be more or less helpful. But two points I have been making are somewhat getting lost in the cycle, and these are the more important points I was trying to make.
First, my original concern was about not just transliterating, but errors in doing so. In “ydaraishy” on f1r, ‘sh’ is assuming that the diacritic and the first c are part of the same letter.
In my examination of the text, that is simply an “error”. That is not a ‘s’, it is a c with a diacritic that belongs most likely to the double-c and there is clearly a space between the diacritic and the first c and I think the parchment defects have made the diacritic longer than what it is in this case (but there a plenty other examples throughout the manuscript that are deniable), and even then it is not touching the c. Whoever is transliterating is making a huge mistake in giving an impression in the transliteration about the ‘s’ and it is misguiding whoever is using that transliteration.
Just a couple of lines below ydaraishy, you have another word, and even when the diacritics are undeniably separate they the c with diacritic is given the 's' assignment as "shoshy"! Why"!? Just because it sounds better?
Of course this is not intentional, and done with good intentions I am sure, but this is not just one case. According to the VM browser, when you search for ‘sh’ 4464 cases are counted when the combination is used as part of a longer string (or word); see the image at the bottom of this post. The error is everywhere. That means an error has been repeated systemically in constructing that transliteration system. That is what makes it a bad system in my view, sorry to say. I would not recommend it, though all of course have a choice to make and in my view that will inevitably lead to slop.
I don’t understand really how anyone, let alone experts, can read a ‘c’ with a diacritic that is clearly a part of a double-c with tops connected as one letter and give it a ‘s’ transliteration. This is not just about an assumption preference. This is about a scientific observation with naked eye. As a result, many who just use that transliteration system without checking the visual origins, will end up studying a Voynich text that does NOT exist as such, because they are studying a particular, in my view verifiably mistaken, transliteration of it.
So, I respectfully disagree with you regarding the benign nature of the choices to be made. I would rather stay with a more cumbersome but more visually accurate transliteration than do something simply because it sounds and pronounces better.
I think you are avoiding this key point I was raising, and my not reminding you of it in my last post may have caused it to be lost in the conversation.
But, then you are avoiding another point I have raised many times, and that is, no matter what transliteration system you use, pronunciation friendly or not, it would be a fundamental mistake to use that system as a basis for making linguistic judgments and drawing such conclusions. From what I have read some have used such statistical studies to claim the language is or not natural, is or not gibberish, is or not this or that language, etc. I think that is a big mistake.
Transliteration systems can be helpful, if they are not obviously mistaken and misleading, to give us some statistics of what exists or not in the text, if visually faithful. However, using them to pass linguistic judgments is a huge mistake and can explain why after decades not much progress has been made in certain studies.
I think what Koen G.’s study of the last page marginalia did, or what JustAnotherTheory found yesterday about the other marginalia go a long way more to help in understanding the Voynich manuscript text. These are good efforts and always bear good results. What JustAnotherTheory found again proves that someone writing in Latin is able to read the text, tentatively speaking.
These seemingly little insights can be path-breaking, but preoccupation with transliteration preferences may take away our attention from them. No one is denying their value (obvious mistakes notwithstanding) in giving us a sense of what exists or not in the Voynich manuscript, but they cannot be used to draw linguistic judgments in a statistical way about the VM text, and their verifiable mistakes should be taken into consideration and red-flagged to readers, not matter how much we appreciate the efforts that have gone into creating them.
Visual studies of the text should be preferred for any convenience transliterations can offer, pronunciation friendly or not.