- Options
- Login
- Desktop Style
Options
Inside the Voynich Network: graph analysis
- Inside the Voynich Network: graph analysis
-
RE: Inside the Voynich Network: graph analysis
Bernd > 02-11-2025, 11:05 PM
Ok, thank you!
I was afraid that the text size of more distinct Voynich 'dialects' would be too small. -
RE: Inside the Voynich Network: graph analysis
quimqu > 03-11-2025, 08:21 AM
(01-11-2025, 06:58 PM)Jorge_Stolfi Wrote: You are not allowed to view links. Register or Login to view.You are sure that the other special characters [µß^~+`'°] are not deleted, remapped, or interpreted as separators at some point?
Hello Jorge,
I have verified and the set of characters of the used text is: ["'", '+', '-', '.', '^', '`', 'a', 'b', 'c', 'd', 'e', 'g', 'h', 'i', 'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'x', 'y', 'z', '~', '°', 'µ', 'ß']
Regards -
RE: Inside the Voynich Network: graph analysis
Jorge_Stolfi > Yesterday, 09:21 AM
New Portuguese and Spanish samples for @Quimqu:
You are not allowed to view links. Register or Login to view.
All texts are derived from the classical 1899 novel "Dom Casmurro" by Brazilian author Machado de Assis.
The "text" folder has the full texts for reference -- more or less as published, will capitals, punctuation, etc. The "full", "head", "tail" have the same texts, but converted to lowercase, with all punctuation (including apostrophe and hyphen) converted to blanks, and with numbers, symbols, and foreign language words replaced by " * "words. Paragraphs are separated by blank lines. Words in each paragraph are separated by blanks and newlines.
The "full" files have all the words of the novel, which are ~65'500 for Portuguese and ~69'000 for Spanish. The "head" and "tail" files are the same, trimmed to the first ~38'000 words and the last ~38'000 tokens, respectively. This token count should roughly match that of the VMS. Thus there is some overlap between each head and the corresponding tail, which is ~10'500 words for Portuguese and ~7'000 for Spanish. The last parag of "head" and the first parag of "tail" may be truncated.
The files are in the Unicode UTF-8 encoding. Besides blank and newline, the characters in the "full", "head", and "tail" files should be
Spanish: abcdefghijklmnopqrstuvwxyz áéíóúñüªº ãç
Portuguese: abcdefghijklmnopqrstuvwxyz áéíóúâêôàãõçüº
In each folder there are four files: "port-orig.txt" uses the original 1899 spelling (apart from capitalization), "port-curr.txt" uses the "modern" spelling (as of ~1999), "port-phon.txt" uses an ersatz phonetic spelling. The files "span-curr.txt" have the Spanish translation of the novel, with (I suppose) current spelling.
The first three files in each folder should have almost 1:1 mapping of the tokens which matches the lexemes. The differences are mostly hyphenated or separate words in one file that are joined in the other, and a few homographs and homophones that are distinguished or merged. The phonetic files, in particular, have the adverbial suffix "mente" as a separate word.
There is no such simple mapping between tokens of Portuguese and Spanish files, even though most lexemes are cognates. The word order and choice of synonyms is often different, etc. I suppose that the Spanish file has more words because the Portuguese contractions are split: "da" <--> "de la" etc.
In any structural word analysis that is independent of a lexical mapping, the three Portuguese samples in each folder should be almost coincident.
And each Portuguese sample should plot close to the corresponding Spanish sample, because, even though they are different languages by different authors (Machado and the translator), the underlying text is the same.
On the other hand, each "head" should map close to the corresponding "tail", because they are different texts but have the same author the same language, the same topic, and the same literary genre and style. They can therefore be used to test how sensitive the analysis is to sampling noise.
The most interesting comparison will be between a Portuguese "head" and a Spanish "tail" (or vice-versa) because they have in common only the topic and genre, and partly the style.
Hope it helps, --stolfi -
RE: Inside the Voynich Network: graph analysis
Trithemius > Yesterday, 07:32 PM
Hey quimqu,
Nice work--
I have also done some graph analysis, but slightly different. In my graph, each node is a folio and an edge connects two pages when they share one or more words that appear at least twice in the entire text and are 4 or more characters long. I then did cosine normalization to prevent pages with more words from dominating so that the connections would reflect shared content rather than just... content/word count.
Anyway what I found was that the manuscript seems to cluster into two clear groups. The left is (I think) pretty clearly herbal A, and the right is the balneological section and the final text-only pages. I think from this graph we can infer that the final pages concern themselves with the balneological matters, rather than the herbal or astrological ones. -
RE: Inside the Voynich Network: graph analysis
quimqu > Yesterday, 08:45 PM
(Yesterday, 07:32 PM)Trithemius Wrote: You are not allowed to view links. Register or Login to view.Hey quimqu,
Nice work--
I have also done some graph analysis, but slightly different. In my graph, each node is a folio and an edge connects two pages when they share one or more words that appear at least twice in the entire text and are 4 or more characters long. I then did cosine normalization to prevent pages with more words from dominating so that the connections would reflect shared content rather than just... content/word count.
Anyway what I found was that the manuscript seems to cluster into two clear groups. The left is (I think) pretty clearly herbal A, and the right is the balneological section and the final text-only pages. I think from this graph we can infer that the final pages concern themselves with the balneological matters, rather than the herbal or astrological ones.
Nice work!
Yes, you can take a look at my thread about You are not allowed to view links. Register or Login to view.. There, it is also clear that the last pages "talk" about balneological, but also about herbal. You can see how.tje automated found topics are.disteibuted.throughout the MS. -
RE: Inside the Voynich Network: graph analysis
Trithemius > Yesterday, 10:11 PM
(Yesterday, 08:45 PM)quimqu Wrote: You are not allowed to view links. Register or Login to view.
(Yesterday, 07:32 PM)Trithemius Wrote: You are not allowed to view links. Register or Login to view.Hey quimqu,
Nice work--
I have also done some graph analysis, but slightly different. In my graph, each node is a folio and an edge connects two pages when they share one or more words that appear at least twice in the entire text and are 4 or more characters long. I then did cosine normalization to prevent pages with more words from dominating so that the connections would reflect shared content rather than just... content/word count.
Anyway what I found was that the manuscript seems to cluster into two clear groups. The left is (I think) pretty clearly herbal A, and the right is the balneological section and the final text-only pages. I think from this graph we can infer that the final pages concern themselves with the balneological matters, rather than the herbal or astrological ones.
Nice work!
Yes, you can take a look at my thread about You are not allowed to view links. Register or Login to view.. There, it is also clear that the last pages "talk" about balneological, but also about herbal. You can see how.tje automated found topics are.disteibuted.throughout the MS.
Thanks! This looks like interesting work, I'll check it out. -
RE: Inside the Voynich Network: graph analysis
Jorge_Stolfi > 5 hours ago
(Yesterday, 07:32 PM)Trithemius Wrote: You are not allowed to view links. Register or Login to view.Anyway what I found was that the manuscript seems to cluster into two clear groups. The left is (I think) pretty clearly herbal A, and the right is the balneological section and the final text-only pages. I think from this graph we can infer that the final pages concern themselves with the balneological matters, rather than the herbal or astrological ones.
Nice! That confirms several other studies that found clustering of pages by section. Like Currier's "languages", and the recent work of @QuimQu. Or this old one of mine:
You are not allowed to view links. Register or Login to view.
You are not allowed to view links. Register or Login to view.
As for the clusters, while Bio and StarrdParags (SPS) overlap in that plot, they are not identical, and can be separated by other statistical criteria.
All the best, --stolfi -
RE: Inside the Voynich Network: graph analysis
quimqu > 31 minutes ago
Here are the plots with Jorge Stolfi's "Dom Casmurro's" and the text in Manx from Ptrarsi.
The four Dom Casmurro versions—old Portuguese, modern Portuguese, phonetic Portuguese, and Spanish—behave almost the same. Their networks have very similar clustering, path length, and degree patterns. That’s expected, since they express the same content and share close grammatical roots. The Spanish version is only slightly denser, likely because Spanish uses more short function words, just as Jorge predicted.
Compared with Lazarillo de Tormes, Dom Casmurro is more uniform and modern in structure. Lazarillo shows greater lexical variety and less repetition, giving it a sparser, more modular network. This can be the difference between current Spanish and older, as the older has a less standardized syntax (16th-century Spanish) and it has a heavier use of subordinate clauses.
The Manx text is clearly different (but very away from the Voynich). It forms a compact network with shorter paths and higher clustering. It seems to be a morphologically rich language.