Re:Translating Akkadian clay tablets with ChatGPT?
by retrac on Tuesday May 16, 2023 @08:59AM
(#35961261)
Not yet. But it's almost certainly possible to use AI in some very useful and important ways in this area.
First, for some background, there's literally hundreds of thousands of tablets that have never been read still, and only a few hundred (and that's a high-end estimate) fully-qualified Assyriologists in the world. We might actually be discovering new tablets faster than they can be read. There's a fascinating story about how Dr. Irving Finkel was wandering through the British Museum archives, picked up a random tablet, only to find a description of the rules on how to play a board game - The Royal Game of Ur - something people had been wondering about for the previous ~50 years after a board and its pieces had been found. Another time he stumbled across what appears to be a predecessor to the Noah ark story, more or less.
Many of the tablets have not even been scanned online or transcribed. Cuneiform is very much a 3D form of writing and it does not photograph very well; at a minimum you want three photographs at different angles. A laser-scanned 3D image is ideal.
Once you have the tablet itself, you want to transcribe it, write down the notation of the tablet, precisely, into something suitable for more easy-handling by us moderns uncomfortable with wedges in clay. And then, finally, you want to try and translate it.
Every step of this is just begging for some automation. I found this a while ago and have been following it: https://github.com/gaigutherz/Akkademia "Akkademia is a tool for automatically transliterating Unicode cuneiform glyphs. It is written in python script and uses HMM, MEMM and BiLSTM neural networks to determine appropriate sign-readings and segmentation."
First, for some background, there's literally hundreds of thousands of tablets that have never been read still, and only a few hundred (and that's a high-end estimate) fully-qualified Assyriologists in the world. We might actually be discovering new tablets faster than they can be read. There's a fascinating story about how Dr. Irving Finkel was wandering through the British Museum archives, picked up a random tablet, only to find a description of the rules on how to play a board game - The Royal Game of Ur - something people had been wondering about for the previous ~50 years after a board and its pieces had been found. Another time he stumbled across what appears to be a predecessor to the Noah ark story, more or less.
Many of the tablets have not even been scanned online or transcribed. Cuneiform is very much a 3D form of writing and it does not photograph very well; at a minimum you want three photographs at different angles. A laser-scanned 3D image is ideal.
Once you have the tablet itself, you want to transcribe it, write down the notation of the tablet, precisely, into something suitable for more easy-handling by us moderns uncomfortable with wedges in clay. And then, finally, you want to try and translate it.
Every step of this is just begging for some automation. I found this a while ago and have been following it: https://github.com/gaigutherz/Akkademia "Akkademia is a tool for automatically transliterating Unicode cuneiform glyphs. It is written in python script and uses HMM, MEMM and BiLSTM neural networks to determine appropriate sign-readings and segmentation."