![]() |
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
A One-Page Ledger Method for Generating Voynich-Like Text - Printable Version +- The Voynich Ninja (https://www.voynich.ninja) +-- Forum: Voynich Research (https://www.voynich.ninja/forum-27.html) +--- Forum: Analysis of the text (https://www.voynich.ninja/forum-41.html) +--- Thread: A One-Page Ledger Method for Generating Voynich-Like Text (/thread-5752.html) |
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
RE: A One-Page Ledger Method for Generating Voynich-Like Text - bi3mw - 31-05-2026 I always welcome constructive criticism. It is indeed a weakness of my rule-based combinatorics that unique words and hapax legomena are left out. One could probably add additional rules to counteract this, but the more rules you implement, the less of a practical method for the author can be derived from them. However, it does look somewhat “synthetic” (“too good”)—it lacks the “human factor.” Unfortunately, that cannot be replicated (afaik). RE: A One-Page Ledger Method for Generating Voynich-Like Text - Dunsel - 31-05-2026 (31-05-2026, 01:21 AM)bi3mw Wrote: You are not allowed to view links. Register or Login to view.Unfortunately, that cannot be replicated (afaik). Well, here's one thing we know for certain. 1 or more humans created this. How and why is up for debate. But if one human was able to create it, then we should be able to replicate it. RE: A One-Page Ledger Method for Generating Voynich-Like Text - rikforto - 31-05-2026 (31-05-2026, 04:02 AM)Dunsel Wrote: You are not allowed to view links. Register or Login to view.But if one human was able to create it, then we should be able to replicate it. Depending on what you mean by "replicate", if the process was not entirely algorithmic and also meaningless, there is a good chance that we will never be able to say with certainty we have the method RE: A One-Page Ledger Method for Generating Voynich-Like Text - Dunsel - 31-05-2026 (31-05-2026, 05:22 AM)rikforto Wrote: You are not allowed to view links. Register or Login to view.Depending on what you mean by "replicate", if the process was not entirely algorithmic and also meaningless, there is a good chance that we will never be able to say with certainty we have the method Here's what I mean by replicate. This is output from my analyzer which is in the repo in the OP. I had GPT convert the output to BBCode to format it for this forum. I'll do 3 more posts with other pages and scribes to show this isn't a 1-off occurence. This shows the source for every word on f1v. Only 1 word doesn't come from You are not allowed to view links. Register or Login to view. and only 2 words are edit distance 2. Note: It only looks at words length >= 3 as length 1 and 2 can be very easily replicated. Interpretation: This table shows that with the exception of 1 token, all other tokens are a product of copy/mutate from You are not allowed to view links. Register or Login to view. or a copy/mutate from a previously written token on f1v. If the "algorithmic method" can be figured out for how they did this, then replication (not duplication) will be possible. And that's what I'm trying to accomplish with my generator. So far, I haven't found an exact method and, as Torsten Timm has pointed out, it's likely a very human method which is not going to be duplicated in code easily. I've stated this before but I'll keep clarifying: I'm not saying this was the method for creating the Voynich. I'm simply asking if it could have been the method. And, if copy/mutate + ledger becomes plausible then then objections that such a process is impossible or even unlikely become much harder to maintain. You are not allowed to view links. Register or Login to view. provenance test Green = ED0 / exact copy after gallows stripping Blue = ED1 after gallows stripping Orange = ED2 after gallows stripping Red = no prior match Core tokens requiring prior source
Same-page local derivations
RE: A One-Page Ledger Method for Generating Voynich-Like Text - Dunsel - 31-05-2026 This is for f20v. It's showing that the sources for the words on this page come from these sheets: q1s1: +17 [CORE] q1s3: +1 [RESIDUE] q1s4: +2 [RESIDUE] q2s2: +1 [RESIDUE] q3s1: +1 [RESIDUE] Residue are tokens that don't come from the core sheet on the initial check and that's how they're listed in the table. RESIDUE TOKENS (NEWLY ADDED BY RESIDUE SHEETS) choraly | stripped choraly | f20v:8:2 -> chodaly | stripped chodaly | f3v:6:4 | q1s3 | ED1 choldy | stripped choldy | f20v:1:7 -> cpholdy | stripped choldy | f4r:8:5 | q1s4 | ED0 shain | stripped shain | f20v:8:1 -> shain | stripped shain | f4r:3:7 | q1s4 | ED0 choiin | stripped choiin | f20v:2:5 -> choiin | stripped choiin | f10v:2:5 | q2s2 | ED0 opydy | stripped oydy | f20v:1:8 -> opydy | stripped oydy | f17r:1:3 | q3s1 | ED0 After a recheck is run, all of those residue tokens have parents on quire 1, sheet 1. RESIDUE CORE-RECHECK RESOLVED ED1: choldy | stripped choldy | f20v:1:7 -> chody | stripped chody | f1v:2:8 | q1s1 RESOLVED ED1: opydy | stripped oydy | f20v:1:8 -> tody | stripped ody | f1v:5:7 | q1s1 RESOLVED ED1: choiin | stripped choiin | f20v:2:5 -> chtaiin | stripped chaiin | f1r:2:5 | q1s1 RESOLVED ED1: shain | stripped shain | f20v:8:1 -> shaiin | stripped shaiin | f1r:22:2 | q1s1 RESOLVED ED2: choraly | stripped choraly | f20v:8:2 -> cthoary | stripped choary | f1r:3:6 | q1s1 Which reduces the retained core source to 1 sheet: quire 1, sheet 1. Again, only length >= 3 were checked. f20v provenance test Green = ED0 / exact copy after gallows stripping Blue = ED1 after gallows stripping Orange = ED2 after gallows stripping Red = no prior match Core tokens requiring prior source
Same-page local derivations
Summary
RE: A One-Page Ledger Method for Generating Voynich-Like Text - Dunsel - 31-05-2026 f44r provenance test Initially, this page reduces to 1 core sheet + 4 more sheets supplying a few extra words. SHEET CLASSIFICATION q1s3: +15 [CORE] q1s2: +4 [SECONDARY] q4s1: +2 [RESIDUE] q6s2: +2 [RESIDUE] q2s4: +2 [RESIDUE] RESIDUE TOKENS (NEWLY ADDED BY RESIDUE SHEETS) dshor | stripped dshor | f44r:6:1 -> dshor | stripped dshor | f25v:5:3 | q4s1 | ED0 dchckhy | stripped dchchy | f44r:9:3 -> dchckhy | stripped dchchy | f25r:4:1 | q4s1 | ED0 ypsholy | stripped ysholy | f44r:1:4 -> yshol | stripped yshol | f42r:8:1 | q6s2 | ED1 kshotol | stripped shool | f44r:3:3 -> shotol | stripped shool | f42r:17:1 | q6s2 | ED0 oracphy | stripped orachy | f44r:1:2 -> torchy | stripped orchy | f13r:8:3 | q2s4 | ED1 oair | stripped oair | f44r:3:1 -> koair | stripped oair | f13v:1:1 | q2s4 | ED0 RESIDUE CORE-RECHECK RESOLVED ED2: oracphy | stripped orachy | f44r:1:2 -> octhy | stripped ochy | f3r:12:4 | q1s3 RESOLVED ED2: ypsholy | stripped ysholy | f44r:1:4 -> shol | stripped shol | f3r:11:2 | q1s3 RESOLVED ED1: oair | stripped oair | f44r:3:1 -> dair | stripped dair | f3r:19:5 | q1s3 RESOLVED ED1: kshotol | stripped shool | f44r:3:3 -> shol | stripped shol | f3r:11:2 | q1s3 RESOLVED ED1: dshor | stripped dshor | f44r:6:1 -> shor | stripped shor | f3r:15:2 | q1s3 RESOLVED ED1: dchckhy | stripped dchchy | f44r:9:3 -> chckhy | stripped chchy | f6r:2:2 | q1s3 After the residue recheck, this page reduces down to 2 sheets with 0 ED2 or unexplained words. q1s3 q1s2 Green = ED0 / exact copy after gallows stripping Blue = ED1 after gallows stripping Orange = ED2 after gallows stripping Red = no prior match / match not in selected minimum set Core tokens requiring prior source
Same-page local derivations
Summary
RE: A One-Page Ledger Method for Generating Voynich-Like Text - Dunsel - 31-05-2026 And here's what happens when you try the same test on a Scribe 2 page. SHEET CLASSIFICATION q5s2: +27 [CORE] q5s1: +3 [RESIDUE] q1s3: +2 [RESIDUE] RESIDUE TOKENS (NEWLY ADDED BY RESIDUE SHEETS) okaldy | stripped oaldy | f43r:2:8 -> okaldy | stripped oaldy | f33r:2:4 | q5s1 | ED0 pshdar | stripped shdar | f43r:10:1 -> tshdar | stripped shdar | f33r:1:1 | q5s1 | ED0 otary | stripped oary | f43r:11:12 -> okary | stripped oary | f33v:8:12 | q5s1 | ED0 otolol | stripped oolol | f43r:11:8 -> otolom | stripped oolom | f3r:18:7 | q1s3 | ED1 shochol | stripped shochol | f43r:13:8 -> shocthol | stripped shochol | f6v:6:2 | q1s3 | ED0 RESIDUE CORE-RECHECK RESOLVED ED1: okaldy | stripped oaldy | f43r:2:8 -> okaly | stripped oaly | f34r:13:5 | q5s2 RESOLVED ED1: pshdar | stripped shdar | f43r:10:1 -> chdar | stripped chdar | f34v:8:7 | q5s2 NOT RESOLVED: otolol | stripped oolol | f43r:11:8 RESOLVED ED1: otary | stripped oary | f43r:11:12 -> okaly | stripped oaly | f34r:13:5 | q5s2 RESOLVED ED2: shochol | stripped shochol | f43r:13:8 -> shotchy | stripped shochy | f34r:8:1 | q5s2 Most words still reduced down to 2 sheets. q5s2 q1s3 There is one word that can't be resolved by ED1 or ED2, "otolol". And, there are 9 words that don't cleanly reduce down to a minimum sheet limit. They do have parents and they can be found, but it's not as easy to reduce as Scribe 1. Exactly what this is saying at the moment, I'm not sure. If Scribe 2 used the same method as Scribe 1, their selection of source sheets was a good bit wider than Scribe 1. Now, all these tests were run with Takahashi. If I instead use ZL transcription for this page, and allow for more initial sheets to be found in the first run, it can be reduced down to 3 sheets q5s2 q5s1 -> added q1s3 with "qosheckhey" being the only word it couldn't resolve to having an ED1 or ED2 parent. Regardless of which transcription is used to locate the source, you can still see that the "proposed" same-page copy mutate process is on par with Scribe 1 pages. f43r provenance test Green = ED0 / exact copy after gallows stripping Blue = ED1 after gallows stripping Orange = ED2 after gallows stripping Red = no prior match / match not in selected minimum set Core tokens requiring prior source
Same-page local derivations
RE: A One-Page Ledger Method for Generating Voynich-Like Text - oshfdk - 31-05-2026 I don't think this shows anything at all. Voynichese is quite regular: the number of common characters is small and their sequences are quite rigid. Most of the vocabulary has many ED1 words. This is a property of the word structure. Let's ask a different question, how many words on You are not allowed to view links. Register or Login to view. have only one ED1 source on f1r, so they can be traced to a single source? Also, I don't understand the rationale for stripping the gallows, this looks like a cheap trick. RE: A One-Page Ledger Method for Generating Voynich-Like Text - rikforto - 31-05-2026 (oshfdk is a little more concise, but I wrote this before seeing their reply and gets at a few more things here) (31-05-2026, 12:57 PM)Dunsel Wrote: You are not allowed to view links. Register or Login to view.Torsten Timm has pointed out, it's likely a very human method which is not going to be duplicated in code easily.One might even say it's not going to replicate in code easily! No one is arguing that there aren't clusters of words in short edit distance, nor that a process which captures that fact can't [edited from "can"] model the text's statistics. It is interesting, and serves as a broad-ranging rebuttal to a number of statistical analyses that had been purported to suggest meaningfulness, that those edit networks plus those constraints have enough information to reconstitute those features. This is a different claim: (31-05-2026, 12:57 PM)Dunsel Wrote: You are not allowed to view links. Register or Login to view.Interpretation: This table shows that with the exception of 1 token, all other tokens are a product of copy/mutate from f1r.When you claim the tokens in fv1 are the product of such and such a process, you are no longer claiming to have modeled them. You are claiming to know as a matter of historical fact what the scribe did. This is an extremely strong claim for the banal reason that it's hard to prove what someone did 600 years ago, but there are several other factors that make inferring from the model hard. These models are in a sense "closely held" to the source text. At risk of broadsiding Torsten, copy-mutate was selected specifically because it reproduces the short edit distances. To the limits of his and his coauthor's argument, that is fine, but it is hardly an emergent property---and without close reading his paper for it, I don't recall him asserting it is. Likewise, the fact that both your models force words through Voynich-like networks of words at rates derived from summary statistics gleaned from the manuscript means that they are likely mutating through spaces of words that have similar properties. The reason why Torsten's paper has force, and why it has the conclusions it does, is that it shows that information is sufficient to produce a Voynich-like text, and so an auto-citation process that has substantially similar properties would be expected to result in a statistically Voynich-like text. Timm and Schinner do not claim the text in the Voynich were produced that way, merely that it is a serious possibility; in fact, they rule out a very literal interpretation of their model! It is entirely possible that their and your models depend enough on patterns that arose from an underlying process interacting with meaningful text that it is reproducing them without reproducing the method. Ruling this is out is very hard. A genuine path forward for these models would be to show that they depend less on the Voynich's summary statistics. Your analysis of the gallows letters, which I do not believe is shared by Timm and Schinner's model, is a case in point. If it could be shown that line start gallows letters were appearing more often at line start because of some simple rule that did not immediately imply the distribution, e.g. that it was emergent, it would lend credence to the idea that these features were dependent on simple rules undergirding copy-mutate rather than the distributions arising in the Voynich for unproved reasons. Your argument (section 4.2 of your paper) is explicitly that these features were inserted based on observations of the manuscript, so you have defined the mutate process to have these properties, and are actually assuming the consequent when you claim to have explained it. In fairness to everyone, it may be the case that there is no simple rule underlying the gallows distribution. It could be that the scribes liked them line start because they looked like capital letters at the start of lines in humanist manuscripts. But this is what I'm driving at. If that's all there is to it, it may be the best we can do is say that their process had a bias towards gallows in line start words and we may not be able to formally separate premise (we observe the bias) from conclusion (it is a product of an arbitrary choice). By the by, if you are in fact "reproducing" the Voynich, failure to incorporate Currier's curve-line system observations and similar analyses seems like fair game to me. It also strikes me as the kind of "observable orthographic structure" you say the model addresses in in 7.6 of your paper. This is largely an aside to the main point here, which is that I don't think you're proving your interpretation, but I'm not clear why statistics about letter bases are not part of the orthographic structures in the Voynich. RE: A One-Page Ledger Method for Generating Voynich-Like Text - Dunsel - 31-05-2026 (31-05-2026, 02:40 PM)oshfdk Wrote: You are not allowed to view links. Register or Login to view.I don't think this shows anything at all. Voynichese is quite regular: the number of common characters is small and their sequences are quite rigid. Most of the vocabulary has many ED1 words. This is a property of the word structure. Let's ask a different question, how many words on You are not allowed to view links. Register or Login to view. have only one ED1 source on f1r, so they can be traced to a single source? Also, I don't understand the rationale for stripping the gallows, this looks like a cheap trick. I think you're focusing on a different question than the one I'm testing. If the claim were simply that many Voynich words have ED1 neighbors, I would agree that this is not very interesting. Given the character inventory and regular word structure, ED1 relationships are expected. The purpose of the provenance test is not to show that an ED1 source exists somewhere in the manuscript. The results are showing that after same-page copy/mutate are removed, the remaining source words repeatedly collapse into a very small number of sheets as the source of those pages. I can easily imagine a scribe with a completed sheet in front of them and using that as the source for the page they're working on. That is what the evidence is pointing to. As for gallows stripping, that was done because gallows are overwhelmingly productive in prefix positions. If gallows behave as a frequent prefix operation rather than as ordinary internal characters, then treating chol, kchol, tchol, etc. as completely unrelated forms obscures rather than reveals the family structure. Whether that assumption is justified is certainly open to discussion, but it was not introduced simply to manufacture matches. Thus far, I do not believe many researchers have seriously considered the possibility that gallows may function primarily as a decorative addition rather than as a core lexical element. I am testing whether they might be. If removing the gallows destroys the structure, then the hypothesis fails. If removing them reveals stable word families, source relationships, and copy/mutate chains that remain consistent across pages, then that is evidence that the gallows may not be carrying the same kind of information as the rest of the word. So far, stripping gallows has not weakened the provenance results. If anything, it has made the underlying relationships easier to see. So, perhaps the cheap trick is not mine to claim. Perhaps gallows is a cheap trick of the scribes' invention. Your question about how many words on You are not allowed to view links. Register or Login to view. have only a single ED1 source is a reasonable one, but it is a different test from the provenance analysis shown above. |