The Voynich Ninja

I want to start with this, because the whole conversation turns on this:

(30-06-2026, 07:30 PM)Jorge_Stolfi Wrote: You are not allowed to view links. Register or Login to view.The Dictator had one written text with 雄鸡 (but in traditional characters) on it, and the Author knew one spoken language, say Cantonese.

I don't misunderstand this. Where we disagree is the nature of that written text and its relationship to the spoken languages. It is entirely appropriate for me to illustrate my point about that with multiple spoken languages, especially since you have yet to identify the language of the VMS. Regardless, though, you are contending there is but one written Chinese language, and so any Chinese language is appropriate to show that is not true. I also used the example languages you used, albeit with a bit more precision, and if you think that's out of bounds, it's because you do not understand your own examples. I don't think you do, and I now understand that's because you do not think that Chinese characters represent words. Until you address my evidence for asserting they are words head on, you will continue to talk past my examples where I've evidenced that that is not true.

Which brings me to this analogy:

(30-06-2026, 07:30 PM)Jorge_Stolfi Wrote: You are not allowed to view links. Register or Login to view.If a flight info site says
  ✈ AA1211 [Plane taking off Icon] 11:00 LAX [plane landing icon] 18:10 SFO
and a text-to-voice software reads it out as
  "flight A A one two one one departure one one zero zero L A X arrival one eight one zero S F O"

These are ideographs because they stand directly for the ideas. There are no conventional readings for them as words and they naturally admit synonyms. In a great many cases they are simply orthographically incorrect, but their use in informal settings can productively take on new words for the idea it stands in for. A person is free to read, "my ✈ is late" as "my plane is late" or "my flight is late" because there is no reading and they must guess the ideas. There is genuinely no phonetic content in ✈. (TTS creates an interesting wrinkle that I can talk about, but I don't think it is clarifying at this juncture.)

As explained at greater length my unanswered post You are not allowed to view links. Register or Login to view., this is not the case with Chinese. The words "飛行" mean flight in Standard Chinese and can only be read with a fixed phonology. Granted, these readings vary between languages and in some cases these characters admit a multiplicity of readings (but rarely more than one at a time in context). The practice of not learning another language's phonology in Chinese is interesting, but it should not obscure that "飛行" cannot be assumed to be the words used for "flight" in every Chinese language. A "reading" is not a freestyle exercise; one must adhere to the conventions of the language at hand, which are based on the characters and the words they represent, not ideography. (Given that most people for most of Chinese history were illiterate, I struggle to imagine how that would have even worked!) One thing I want to stress again is that none of this is foreign to European languages, it is only more pervasive in Chinese-derived orthographies. Latin admits a great many phonologies through regional pronunciation while still being fundamentally Latin; there are homonymous false friends between various languages, such as "actual" in Spanish and English. There are genuine transference benefits of this for literate speakers of Northeast Asian languages, but they do not come close to full mutual intelligibility without extensive training.

I was not as pedantic as I could have been in my post You are not allowed to view links. Register or Login to view. because I had not yet realized that you did not understand Chinese languages are written in words, but it does clearly show the substance of this distinction, and the dictionary entries it is based on back the premise Chinese languages are written in words. I trace how different phrases for "rooster" give rise to different spoken terms when read character-by-character, but that not all those readings produce attested phrases in every language. Of particular interest, the Classical Chinese phrase for Rooster does not exist in the prestige dialect of Cantonese. This is not an issue of pronunciation as the reading exists; it is the phrase that does not. Nor an issue that I do not understand that The Dictator was supposed to be giving character readings from the same language as The Author; I have long conceded that your method produces proper readings based on Cantonese. My contention is that Chinese is written in words, and words are not the same between distinct languages, and that Classical Chinese is one of those languages.

I finally broke down and tried out this exercise with Google's LLM, as insulting as it is to place a thoughtless machine with a random number generator in front of the great many worthwhile human beings who contributed to the dictionaries, academic work, and popular discussion that have been cited above. I got very different results from you, but it doesn't make sense to discuss that until we are on the same page about whether or not Chinese characters represent words because my best explanation---besides the undeniable fact that LLMs are a black boxes that do not follow understood principles---requires us to be on the same page about that.

By the way, you have the answers to those "simple questions" you mention in your post, and until you deal directly with my evidence and observation that gai1 is not the head of the phrase that means "rooster" in Cantonese, you will not convince me you are following my argument and looking at the attendant evidence. This is true whether "gai1" is written <gai1>, <鸡>, or <雞>; if you the statement before the semicolon is nonsense under your understanding of how Chinese is written, that must be resolved before my points will make sense. (It still makes sense on its own terms, but if you convince me that that is wrong, it will moot the point; at any rate, it's the heart of our disagreement.)

Some things that will not bring us closer together, regardless of which of us is correct:

• Repeating your conception of how the dictation worked; we do not agree about the nature of the languages in play, so simply telling me which languages you think were being written and read will not resolve that I don't hold Classical Chinese to be a written register of Cantonese. I understand that if it is, a lot of my objections go away.
  
• More examples of ideographs. I understand how ideographs work, and how you believe they apply to Chinese. You need to explain to me why lexicographic sources do not bear out your analogy. (Also, why my Japanese teachers used to take points off when I confused readings and synonymous kanji. My Chinese is quite poor, though I have some basic familiarity with Classical and Standard as a objects of linguistic study, but I was at one point quite used to reading words in Chinese scripts.)
  
• More examples from Google "AI". If we do not agree about what the LLM is outputting (words or pictures), there is no point discussing why we might also be getting different output. This is true irrespective of the impact it's having on our prompting, but I strongly suspect that it is directly and indirectly relevant.
  

(01-07-2026, 02:29 PM)rikforto Wrote: You are not allowed to view links. Register or Login to view.I was not as pedantic as I could have been in my post You are not allowed to view links. Register or Login to view. because I had not yet realized that you did not understand Chinese languages are written in words, but it does clearly show the substance of this distinction, and the dictionary entries it is based on back the premise Chinese languages are written in words. I trace how different phrases for "rooster" give rise to different spoken terms when read character-by-character, but that not all those readings produce attested phrases in every language. Of particular interest, the Classical Chinese phrase for Rooster does not exist in the prestige dialect of Cantonese. This is not an issue of pronunciation as the reading exists; it is the phrase that does not. Nor an issue that I do not understand that The Dictator was supposed to be giving character readings from the same language as The Author; I have long conceded that your method produces proper readings based on Cantonese. My contention is that Chinese is written in words, and words are not the same between distinct languages, and that Classical Chinese is one of those languages.

There are even more points to add:
- Not all texts from China of the time we should consider were written in a script we nowadays would identify as Chinese, some were, to name one, based on the mongolian script. 

- Next, there are huge differences between Wényán, the script a local copy of your book would uttermost probably have been in, and Báihuàwén, the script your local reader would have expected, at least in most parts of China, see my first remark. So much has changed: word meanings, order of strokes, (often tonality as well, just to add). He could have been able to understand Wényán, if he was well educated, but he would then struggle to find words for it. Imagine yourself reading a menu in a restaurant in, let's say, Albania. You will certainly have the idea what "Spageti" means, get a good or not so good guess on "Mish viçi me rrikë" (as you may know other languages as well). But Tavë Elbasani? You wouldn't even know how to pronounce it. And now add the assumption parts of the menu are written in the traditional Todhri alphabet. The pictures on the menu don't help a lot.  

- And finally, true, Wényán was mostly about words. One Sign, one word, that was where Chinese started - almost a strict rule with only few assumptions. But this changed partly with classical chinese, this is why there where rhyme dictionaries, helping with both syllables (Fǎnqiè) - with do not work the same allover the country, as the regional uses multifurcated. And Pǔtōnghuà, the language your AI tool uses for the sake of not knowing anything about classic chinese texts, is again totally different. Most words of today's Chinese are made of two, sometimes even three or more signs. And a lot of words from now appear to be false friends, having changed their meanings, not to say their pronounciations and a lot more. Even for these differences alone one may say that your attemps are like translating Latin using a handbook of contemporary Castellano.  I strongly recommend you to skip the usage if AI here or setup a dedicated, owned LLM for precisely your purposes. Every new website Google founds would change your outcome otherwise.


Cheers,

Doro

(02-07-2026, 10:17 AM)DorotheeM Wrote: You are not allowed to view links. Register or Login to view.get a good or not so good guess on "Mish viçi me rrikë"

My guess was badly wrong: "mixed fishes with rice" 

(30-06-2026, 05:08 PM)rikforto Wrote: You are not allowed to view links. Register or Login to view.Eggyk, if Jorge retreats to your position I will address it in more detail,

The entire origin of this specific theory (and the reason a dictation backstory has been proposed) is based on the supposed match in the relative positions of daiin with the relative positions of the word "use" in the SBJ. To 'retreat' in this position would be to retreat to the initial topic imo. 

It's entirely possible that SPS is a voynichese version of the SBJ, AND also possible that the dictation scenario is wrong/impossible. Shouldn't the focus be on confirming if the texts match before determining exactly how it was written? This detailed, interesting and lengthy debate is focusing on the 'how', when we still don't know the if. 

Even if the dictation scenario were to be entirely falsified, that would not falsify the foundation of the theory. If the underlying theory of a match is falsified, the discussion on how it came to be isn't necessary.

Hi, Jorge!
The discussions in this thread must be very interesting and informative, but they're so long that I'm completely lost. 
I'm eagerly awaiting your results, because I firmly believe that even if the original was in Chinese, our text would be a translation, not a transcription.

P.S. I have another very simple question: where is the link to your current research? 
The link in your profile on this forum doesn't lead to anything I'm interested in. 
Could you please add the updated link to your profile or even your signature? 
Thank you in advance.

(02-07-2026, 10:50 AM)eggyk Wrote: You are not allowed to view links. Register or Login to view.The entire origin of this specific theory (and the reason a dictation backstory has been proposed) is based on the supposed match in the relative positions of daiin with the relative positions of the word "use" in the SBJ. To 'retreat' in this position would be to retreat to the initial topic imo. 

I have said at length that that "match" is no such thing. In short, if you have to create a new critical edition of the Shennong Bencao Jing (where you delete the most common verb???), allow the edited characters to match 120 different Voynichese strings (not even orthographic words, conventionally understood), and only get a statistical match for a third of the entries you test, I don't think we can reject the null hypothesis that those cribs are spurious. I've said this, and other people have said most of this.

But Jorge is the one resting part of his argument on the idea that Chinese people are illiterate and just communicate in pictures, and I think it is worth correcting the record on that for its own sake, for reasons that should be plain from that formulation. But it is relevant to this thread, as he thinks it is entirely plausible that The Author could read all Chinese texts (which under this conception would not be "texts" as such) but for knowing the readings, has based his You are not allowed to view links. Register or Login to view. on those assumptions, has assumed that Chinese has no grammar (!!!) that would get in the way of doing that, and when asked about the plausibility of the text being in Chinese goes back to the dictation scenario. These problems would remain, albeit with less urgency, if I found the match convincing, but as things stand, I think they go to what a longshot this "match" is for exactly the reasons that Jorge has argued they are not problems

(02-07-2026, 02:53 PM)rikforto Wrote: You are not allowed to view links. Register or Login to view.I have said at length that that "match" is no such thing. In short, if you have to create a new critical edition of the Shennong Bencao Jing (where you delete the most common verb???), allow the edited characters to match 120 different Voynichese strings (not even orthographic words, conventionally understood), and only get a statistical match for a third of the entries you test, I don't think we can reject the null hypothesis that those cribs are spurious. I've said this, and other people have said most of this.

I don't really disagree with you on that, for the record. And i'm not trying to stop you from discussing these problems. 

The reason I posted the graphic was to show that two different versions of the same text, with completely different grammar (i assume) and different lengths have "matches" that look similar in accuracy to the "matches" Jorge first presented. If I had shown that graphic, but hypothetically presented it as voynichese vs classical chinese, and then said it's evidence of an Voynichese-"Art of War" equivalency, I assume the response would be very critical. The point is, though, that those matches really are matches, even though they look spurious. 

(02-07-2026, 02:53 PM)rikforto Wrote: You are not allowed to view links. Register or Login to view.These problems would remain, albeit with less urgency, if I found the match convincing, but as things stand, I think they go to what a longshot this "match" is for exactly the reasons that Jorge has argued they are not problems

This is mostly what I was trying to say. However, my personal interest would be working out whether or not a convincing match is there before getting into the other issues.

(02-07-2026, 05:26 PM)eggyk Wrote: You are not allowed to view links. Register or Login to view.The reason I posted the graphic was to show that two different versions of the same text, with completely different grammar (i assume) and different lengths have "matches" that look similar in accuracy to the "matches" Jorge first presented. If I had shown that graphic, but hypothetically presented it as voynichese vs classical chinese, and then said it's evidence of an Voynichese-"Art of War" equivalency, I assume the response would be very critical. The point is, though, that those matches really are matches, even though they look spurious. 

Yes, my explicit position has been for quite some time this is not a robust result, and I think that graphic indicates how much we should not expect a provable match from this approach. And I think it's worth remembering that if the text is in translation, it is not the text that Jorge has identified, and that bears on both the claim it's been identified and how to proceed. Let's grant that this works as an identification method. Is the solution to the differences deleting clauses from the original Art of War? Removing a bunch of the verbs? (!!!) Insisting that the Art of War need not be translated because there is no language in the original to be translated? No, and so this would still bear on Jorge's methodology and claims. I would prefer to have the conversation on those terms because I think it is likely to reveal problems in accepting the match as made. But the current barrier to that discussion is that there is no recognition that the second line of your graphic even exists, let alone what the consequences of recognizing that would be.

(02-07-2026, 10:17 AM)DorotheeM Wrote: You are not allowed to view links. Register or Login to view.- Not all texts from China of the time we should consider were written in a script we nowadays would identify as Chinese, some were, to name one, based on the mongolian script.

OK,but that is not relevant.  The source of the Shennong Bencao Jing (SBJ), that the Author may have had access to, can only have been one of the printed medical encyclopedias that quoted it in its entirety, like the Zenghe Bencao (ZHB, first composed in ~1080 CE).  These were in Chinese characters, not Mongolian ones.

Quote:Next, there are huge differences between Wényán, the script a local copy of your book would uttermost probably have been in, and Báihuàwén, the script your local reader would have expected

The text that the Dictator had to read was definitely in Wényán ("Classical written Chinese"), and a very old form at that (apparently 200 CE or earlier).  

Hopefully the Dictator was a well-educated medical doctor who knew all the medical hanzi, and was used to reading medical books in Wényán.   But that assumption is not really necessary, because neither the Dictator nor the Author needed to know the syntax of the text's language.   Or even the meaning of the text.  Anyway, the SBJ has practically no complicated sentences: each entry is mostly a list of diseases or benefits of the drug, and the structure of each item (2-4 hanzi long) is almost always noun-adjective or verb-object. The differences between Wényán syntax and the syntax of the local vernacular would have been the least of the obstacles to the Author's understanding.

Quote:He could have been able to understand Wényán, if he was well educated, but he would then struggle to find words for it.

Maybe he did not know how to read some medical-related hanzi in the local vernacular (let's call it "Cantonese" for convenience).  Maybe he knew only how to read it in some other language, that people in that area used when reading Wényán.  In that case he may have used that reading.  The Author would not have understood the term either way; but would still have recorded it in the phonetic script, and perhaps asked for clarification later.

Quote:let's say, Albania.

A more pertinent analogy would be the Albanian guy at the Albanian pizzeria dictating the menu over the phone to a tourist at the hotel, who has only a limited knowledge of spoken Albanian. This tourist would write the words down phonetically, as best as he could, hoping that he could identify "presh" and "borzeelock" and other unknown words later in his phrase book.  

Quote:I strongly recommend you to skip the usage if AI here or setup a dedicated, owned LLM for precisely your purposes. Every new website Google founds would change your outcome otherwise.

Since I can't read or speak any Chinese language, I cannot avoid the lalamos for extracting and translating the SBJ text.  Fortunately, every hanzi of the SBJ has been discussed in hundreds of scholarly papers and books, so they don't lack reliable sources (which Google AI cites).  And I routinely check the output of one lalamo against another, until they agree.  

Anyway, the impact of eventual lalamo errors is limited, because at this point I do not need to know what condition 贲豚 is exactly, or how 主 was pronounced by the Dictator.  The key facts I depend on are that

1. Most of the time, one hanzi in the SBJ text (as quoted in the ZHB) corresponds to ~5 EVA letters in the VMS, with surprisingly little variation;
   
2. Overall, the lengths of the SBJ recipes match fairly well those of the parags of the SPS, accounting for the above ratio;
   
3. Each hanzi 主 in the SBJ almost always maps to a daiin in the VMS, at the expected place; except for occasional variations like laiin, kaiin, dain that look like errors by the VMS Scribe due to confusion of letters with similar shape. And the same occurs for a few other characters.
   

All the best, --stolfi

(02-07-2026, 12:10 PM)Ruby Novacna Wrote: You are not allowed to view links. Register or Login to view.The discussions in this thread must be very interesting and informative, but they're so long that I'm completely lost. I'm eagerly awaiting your results, because I firmly believe that even if the original was in Chinese, our text would be a translation, not a transcription.

Well, from what I have seen so far, if it is a translation, it must be very close to literal, and into a monosyllabic language: mostly 1 hanzi = 1 syllable = ~1 Voynichese word (modulo missing and bogus spaces) = ~5 EVA letters (not counting spaces).  The locations of the "cribs" are too precise for it to be a free translation.

But I am not sure of that yet.

Quote:I have another very simple question: where is the link to your current research? The link in your profile on this forum doesn't lead to anything I'm interested in.

Shame on me, I should update my website at the university.  It is only ~30 years out of date...

But for the SBJ=SPS question, besides the TeX paper, I have You are not allowed to view links. Register or Login to view. that record my attempts to find the SPS parag that matches each individual SBJ recipe.  This index page lists those recipes that I have tried so far (39 of them, still 327 to go...).  You are not allowed to view links. Register or Login to view. is generated mechanically from five files: the text of the recipe in You are not allowed to view links. Register or Login to view., You are not allowed to view links. Register or Login to view., and You are not allowed to view links. Register or Login to view., my You are not allowed to view links. Register or Login to view. (the same for all recipes), and a You are not allowed to view links. Register or Login to view. (specific to each entry) that processes those four files and outputs an You are not allowed to view links. Register or Login to view.. 

Currently I am extracting the hanzi version of each recipe from You are not allowed to view links. Register or Login to view. obtained from an edition of the Zhenghe Bencao (ZHB) encyclopedia printed in the 1700s.  Unfortunately that file does not clearly mark out the parts that are  quotes of the SBJ.  They would have been printed white-on-black in an edition available in 1400, but later editions dispensed with that clue and printed everything black-on-white.  Thus I must rely on the help of lalamos to identify the SBJ parts.  They have access to tons of scholarly papers and books about the SBJ, and probably some critical editions of it, so they should "know" which parts are original. But the lalamos often contradict themselves and each other, and it takes about one hour to get a text of the recipe that I can barely trust.  

The Mandarin pinyin file is not necessary; it is used only to make the html report more accessible to those (like me) who can't read Chinese characters but (unlike me) may understand spoken Mandarin.  The English translation is necessary to identify parts of the recipe that may have been omitted by the Author.  Some parts are consistently omitted (the [flavor], [another name],and [provenance] fields), but sometimes the recipe includes other bits that the author may or many not have skipped too.  Like the comment "amber is a divine substance" after the claim that chicken eggs can be turned into amber, or the qualifier "(for women)" before gynecological conditions or benefits.  When I notice such an item in the English text, I must tell my scripts to try to find a match both with and without that part, then check the results and try decide whether the part was or was not actually omitted.

The "good part of the SPS" file is the text of the Starred Parags section from line f103.1 to line f116r.30, thus excluding the "unstarred" block at the end; minus some parts where there were two or more parags lumped together and I could not decide where the division was.  In particular, it excludes three large blocks on f108r, f108v, and You are not allowed to view links. Register or Login to view. where the Scribe apparently ignored parag breaks and mashed 10-12 parags in a continuous block of text.  Thus it has onlt 243 parags instead of the 330 of the full transcription (and of the ~365 that presumably existed before the loss of the central bifolio).  The transcription is a new one that I made from scratch and checked against Rene's; it differs in many but small details.  That particular file has one parag per line, with no spaces [-,.], and with all weirdos and funny ligatures turned into '?'.

However be warned that those pages are still rough drafts, not fit for wide public distribution.  Some (with greenish background) use an incorrect hanzi text of the recipe from another SBJ file that had many small but sometimes critical differences from the ZHB version.  Others (bluish background) use the ZHB version for the matching, but the  "Discussion" section is still nonsensical or obsolete because it was written when I was still using the wrong version.  Thus do not bother to point out errors in those pages yet...

All the best, --stolfi