To summarise the inferred times, it would be:
* 3 weeks up to a month to generate gibberish
* up to around a season to encipher some text (assuming you already had the underlying text)
* when copying faithfully, I found an estimate of around 2 pages a day for a good scribe; that comes out to around 5 months or maybe half a year. This would be assuming the scribe knew the language, though. A faithful rendition if you don't might take longer. (this also assumes an existing draft). This rough speed is somewhat confirmed by a comment on a post my friend found of someone who does calligraphy using archaic methods for fun.
So enciphering takes 3-4 times as long as gibberish, and then a faithful copy takes a little less than double of that, or 6-8 times longer than gibberish.
Edit: sorry, this should go to some of the "gibberish" threads.
(13-03-2026, 04:31 PM)dexdex Wrote: You are not allowed to view links. Register or Login to view.The methods devised by participants of g&b experiment can match Voynichese quite well (and seemingly better than any proposed languages),
By what criteria, and who measured them?
A simple first-order Makov chain will produce gibberish with most of the same vital statistics as the Shakespeare corpus -- Zipf plot, entropy per word, n-gram frequencies, word pair frequencies. Someone who does not know English would say that the synthetic text matches Shakespeare "quite well". It would match it better than Agatha Christie, Nicholas Culpeper, or any other natural language text would.
Would that suggest that Hamlet may be just gibberish? That it cannot be a natural language?
The obvious implementation of a zero-order word-based Markov would use a single table of word probabilities. For a first-order word Markov, one would need a separate table for each word X in the lexicon, that gives the frequencies of each word Y following word X. That would be quite large if it was complete; but if the frequencies are computed from a fixed "seed" text of N words, all those tables together will have at most N-1 entries.
But then, instead of a table, the seed text itself can be used to represent those frequency distributions implicitly. To generate the next word Y, you take the previous generated word X, pick a random occurrence of X in the seed text, and copy the following word. This trick (which works for higher-order Markov chains as well) makes it possible to chose the next outout word without generating random numbers with unequal probabilities, which a scribe in the 1400s would hardly be able to do.
Thorsten&Timm's "self-copying" method seems to be quite similar to this implementation of a Markov chain. Except that the generated text itself is continuously appended to the seed text, and a separate (non-trivial!) algorithm is used to occasionally create new words by mutation of previous ones. Ah, and also a time-travel routine is called initially to fetch the last page of the output (page f116r) from the future to use as the initial seed text (page f0v), which is discarded at the end.
All the best, --stolfi
(13-03-2026, 08:48 PM)Jorge_Stolfi Wrote: You are not allowed to view links. Register or Login to view. (13-03-2026, 04:31 PM)dexdex Wrote: You are not allowed to view links. Register or Login to view.The methods devised by participants of g&b experiment can match Voynichese quite well (and seemingly better than any proposed languages),
By what criteria, and who measured them?
A simple first-order Makov chain will produce gibberish with most of the same vital statistics as the Shakespeare corpus -- Zipf plot, entropy per word, n-gram frequencies, word pair frequencies. Someone who does not know English would say that the synthetic text matches Shakespeare "quite well". It would match it better than Agatha Christie, Nicholas Culpeper, or any other natural language text would.
Would that suggest that Hamlet may be just gibberish? That it cannot be a natural language?
The obvious implementation of a zero-order word-based Markov would use a single table of word probabilities. For a first-order word Markov, one would need a separate table for each word X in the lexicon, that gives the frequencies of each word Y following word X. That would be quite large if it was complete; but if the frequencies are computed from a fixed "seed" text of N words, all those tables together will have at most N-1 entries.
But then, instead of a table, the seed text itself can be used to represent those frequency distributions implicitly. To generate the next word Y, you take the previous generated word X, pick a random occurrence of X in the seed text, and copy the following word. This trick (which works for higher-order Markov chains as well) makes it possible to chose the next outout word without generating random numbers with unequal probabilities, which a scribe in the 1400s would hardly be able to do.
Thorsten&Timm's "self-copying" method seems to be quite similar to this implementation of a Markov chain. Except that the generated text itself is continuously appended to the seed text, and a separate (non-trivial!) algorithm is used to occasionally create new words by mutation of previous ones. Ah, and also a time-travel routine is called initially to fetch the last page of the output (page f116r) from the future to use as the initial seed text (page f0v), which is discarded at the end.
All the best, --stolfi
I think this is straying from the topic of this thread, but this is the paper: You are not allowed to view links.
Register or
Login to view.
(13-03-2026, 08:42 PM)dexdex Wrote: You are not allowed to view links. Register or Login to view.I found an estimate of around 2 pages a day for a good scribe
I guess that this estimate is for a good scribe writing high quality text with neat uniform "font", ornate letters, etc. Not just a faithful copy.
- You are not allowed to view links. Register or Login to view. seems to have been written directly by the author without a draft, "from brain to paper".
- You are not allowed to view links. Register or Login to view. was written by a professional scribe, probably in a "book factory", copying literally from some earlier copy. It had to be a faithful copy, but the scribe made many errors that were corrected, presumably by a proofreading scribe. I suppose that it is a typical commercial-quality product, that a scholar would find good enough.
- You are not allowed to view links. Register or Login to view. is a higher-quality product where obviously a neat and uniform "font" was essential and visible corrections would not be tolerated. I guess that it was meant for the luxury book market, for rich clients to show off. A Vuitton book.
Apart from the ornate initials (which are practically lacking in the VMS), and the proofreading, I would guess that book 2 took less than 1 hour per page to write, and book 3 did not take more than twice that. The scribe of 3 just had to write a bit more slowly to make sure that the strokes would be all the proper size and shape; while the scribe of 2 could let his hand move "semi-autonomously".
All the best, --stolfi
(13-03-2026, 09:19 PM)Jorge_Stolfi Wrote: You are not allowed to view links. Register or Login to view.I guess that this estimate is for a good scribe writing high quality text with neat uniform "font", ornate letters, etc. Not just a faithful copy.
While it's certainly not just 'scribbling,' this estimate is from the era when lay education was already a thing, necessitating more books being copied. The copying hand was a little more lax than ornate books, but reproducing text readably and with decent spacing still takes time even if it doesn't need to be perfect. A book reproducing extensive calligraphy and very elaborate letters, exhibiting neatness, would take much longer. There are reports of one such copying where the contracted scribe had to produce folios continuously for the client: the rate came out to about a page a day, so a year for a tome of Voynich size (well, 9+ months, but if it takes this long you will run into shorter days, the cold season, and illnesses interrupting your work, even in Italy, so a year is a more realistic estimate).
The Voynich is assuredly not of this quality. Going further, there is no line spacing, so one might infer it should be a 'quicker' copy, but the handwriting is not of 'shorthand' quality inasmuch as we can even guess at it, and the drawings need to be reproduced and text wrapped around them, so it can't be freely written in the way a quick copy would be. I would say a typical copy's time of production is a decent guess.
(13-03-2026, 09:27 PM)dexdex Wrote: You are not allowed to view links. Register or Login to view.The Voynich is assuredly not of this quality. Going further, there is no line spacing, so one might infer it should be a 'quicker' copy, but the handwriting is not of 'shorthand' quality inasmuch as we can even guess at it, and the drawings need to be reproduced and text wrapped around them, so it can't be freely written in the way a quick copy would be. I would say a typical copy's time of production is a decent guess.
You can get an estimate yourself by printing one Herbal page at true scale (23.5 cm tall) to use as a draft, then reproducing the drawing and copying a few lines by hand with a sharp felt-tip pen, clocking the time it takes for each task. Then you can do the math...
All the best, --stolfi
Undoubtedly there is plenty of historical evidence about the speed of professional scribes, but I don't have this.
My feeling is that the 'committing to paper' part is perhaps not the largest part of the entire creation process.
Also, if the MS is a fair copy, one has to include the time both of creating the draft and creating the copy.
(13-03-2026, 11:29 PM)ReneZ Wrote: You are not allowed to view links. Register or Login to view.My feeling is that the 'committing to paper' part is perhaps not the largest part of the entire creation process.
Also, if the MS is a fair copy, one has to include the time both of creating the draft and creating the copy.
If it contains meaning, that's quite probable, though it could be an encipherment of a plaintext book that no longer exists. Then the author and scribes are technically different, and their motives can be separate and so the time investment may be 'worth it' from each standpoint when it might not be worth writing a book only to encipher it...
Without a decipherment, speculating on the production of the draft is even more uncertain. Scribing we at least can do some guesses.
Probably having such long text with meaning is a year's worth of labor, I would guess. But poor quality babble with linguistic meaning but full of bullcrap could be done faster or it could be a decade's worth of work by some herbalist, no way to tell.