I'm interested to know about the approaches that you tried with the ML, and then decided to not use. In practice, the options are so many. How did you come up with the final approach - and was there a systematic way to decide which options to go for?
replyOutstanding work! I've participated in the challenge, but didn't get far. One of the questions I had at the time was - if I'm going to use ML to detect ink, could it invent hallucinated letters, or even parts of text, and how to prevent that?
replyYes, it's quite possible for ML to hallucinate ink, though it is on a much more local scale, like predicting a slightly longer stroke, filling in more of a character than is actually in the data, etc. Perhaps enough to change a reading of a character or show where ink isnt. It is difficult for ink detection to hallucinate grammatical and idiomatic greek and latin.
replyWhat is the input to the ML algorithm? Does it know the surrounding context so that it has a chance to deduce "if this stroke is slightly longer then the end result will be idiomatic greek and latin"?
replyThe input is 3d chunks of reconstructed CT data from our scans. I can't remember the specifics but maybe enough voxels for .5mm^3 at a time or so? They're all available for free from https://registry.opendata.aws/vesuvius-challenge-herculaneum... . Our trained models are all available at https://huggingface.co/scrollprize
replyNot all machine learning is generative AI.
replyTrue but like regular document scanning software there can be errors in detection.
replyJust as with redacted documents (consistently blocked terms) or bad OCR jobs (wrong or missing characters), even if only a certain percentage comes out unmangled it is more readable than having no data at all.
replyA stable base corpus and some dynamic programming will allow you to clean up the remainder[0].
Absolutely incredible work. This is one of the most amazing news articles I’ve encountered in decades. Congratulations team!
replyWhat are the wildest, most exciting but plausible things that might be discovered in these documents?
replyI am not a papyrologist or a classicist, rather I'm a computer scientist, so my expertise is unfortunately not in _what_ the scrolls say, rather how we get there. That being said I think and hope that there will be a trove of things that has no known provenance at all, completely lost works that elude the public memory.
replyYour response reminds me of Nigel Richards :)
replyhttps://en.wikipedia.org/wiki/Nigel_Richards
Congratulations, and thank-you!
Well what were your first thoughts when you decoded the script, besides the obvious Eureka, after making some sense of the texts?
replyProbably a lot more texts of Epicurean philosophy and not a whole lot else unfortunately according to my papyrologist friend.
replyThat's what was thought, but maybe not -- only one of the three so far looks Epicurean, which is not what was expected. Maybe it's a fluke, but historians are buzzing a bit about whether it might be broader than expected.
replyWhy would Epicurean philosophy be unfortunate?
replyI was under the impression that there was almost nothing left of that school of thought, and that it’s writings had been destroyed.
What would you like to have instead?
The unfortunate part is the lack of anything else therein, not that it's Epicurean philosophy.
replyThe Jewish Talmud uses Epicurus's name as a term meaning "heretic".
replyHere's a list. The scrolls are from a library that burned in 79 AD.
replyWoah there was a lost Homer epic comedy about a bumbling fool named Margites?
replyHow do get to do that? As in what did you study to get the prerequisite knowledge, and how did you find this particular job? When I see interesting jobs I'm anyways curious what path lead there
replyI am a computer scientist. I studied CS in university, worked in the semiconductor industry for a while, got started as a participant in the challenge aspect of the Vesuivus Challenge. They were hiring, I sent in an application, interviewed, and was offered the job.
replyThat last sentence is so perfect, like my dad answering the question of how he lost weight. "I ate less and exercised more."
replyAmazing work, fantastic!
replydeleted
replyNo questions, but I just want to say this is really exciting work!
replyGiven the current rate of progress, how long do you think it will take to decipher the entire collection?
replyThat's a tough one to give a strong estimate of. Some scrolls are easier or harder to unwrap and read for a multitude of different reasons, mostly due to how damaged the scroll was in the eruption, and how easy or not the ink is to read. IIRC from what we've scanned of the herculaneum collection, none of the ink is easily visible via spectrum alone, so we have to use a lot of ML and physically based rendering techniques to be able to find ink. That also requires unwrapping and segmentation _before_ any ink detection.
replyFor iron gall ink with high enough iron concentration, the ink stands out in the xray volume through simply masking off low values, such as was shown in our campfire scroll experiment a few years ago. No herculaneum scrolls show similar ink.
Do you think this particular scroll is easier or harder to read that the others will be? Or about average?
replyPherc1667 was quite small and just so happened to have readable ink, so it was easier than I expect most others to be.
replyDo we known what ink is used?
replyMost of the evidence so far points towards carbon based ink. I am not sure if any of the scrolls we have scanned show strong evidence of iron gall based ink. I know that there are different types and preparation methods for different carbon based inks, but I do not know if it is possible to determine which kind(s) were used solely from inspecting the xrays.
replyI am, though, not a papyrologist, so historical ink making, preparation, and usage are not my field.
Thanks!
replythis is überragend.
by many means!
replyDid anyone on the team come from a non-science, non-math, non-academia background? Did anyone working on this just teach themselves and start contributing?
replyYes. Sean, who was a co-winner of the 2024 prize, IIRC has no formal background in ML, computer science, AI, etc. He is one of our core researchers and the most productive team member.
replyI've been on the Discord for a couple of years now, and poking around with submissions as well. Sean and the entire team deserve so much praise for all of this work.
replyIt's easy to just read about the breakthrough and see it as one neat, linear line to get there, and hard to comprehend the hours, months and years that so many spent to get there. Big congrats to you, Sean, Nat and the entire team!
Are the fragments destroyed in ‘69 and ‘80 available to be read similarly? Or were they disposed of?
replyI am unaware of those fragments in particular. Though we have scanned a dozen or so fragments, mostly to help guide ink detection, since the ink in them is often more visible in visible and/or near IR light, but can be hard to impossible to detect in the xray spectrum.
replyI don't have any questions, just a comment.
replyYou have a potential to rewrite the history of European Antiquity quite substantially. The Herculaneum set of scrolls is enormous and must contain a lot of hitherto unknown.
That comes with a set of peculiar risks. Once your work starts producing something that contradicts previous work of Very Important People, they will lobby to stop you. Be prepared for that.
Science should be neutral and always value new evidence. Scientists as humans are unfortunately subject to all sorts of passions.