If you want to skip the journalistic fluffery and go to the technicals here's the paper:<p>Reading Akkadian cuneiform using natural language processing (NLP):
<a href="https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0240511" rel="nofollow">https://journals.plos.org/plosone/article?id=10.1371/journal...</a>
Id love to see ML take on Codex Seraphinianus. Even if the answers are just hallucinations, they'd fit thematically.<p><a href="https://en.m.wikipedia.org/wiki/Codex_Seraphinianus" rel="nofollow">https://en.m.wikipedia.org/wiki/Codex_Seraphinianus</a>
New Fragmentarium Website: <a href="https://fragmentarium.ms/" rel="nofollow">https://fragmentarium.ms/</a><p>Old: <a href="https://www.ebl.lmu.de/fragmentarium" rel="nofollow">https://www.ebl.lmu.de/fragmentarium</a><p>Related papers:<p>- <a href="https://aclanthology.org/2024.lrec-main.1197.pdf" rel="nofollow">https://aclanthology.org/2024.lrec-main.1197.pdf</a><p>- <a href="https://openreview.net/pdf?id=z6ZGKexu8un" rel="nofollow">https://openreview.net/pdf?id=z6ZGKexu8un</a><p>NLP-enabled string matching. I wish there were more details about _how_ they did it in the NYT article since that would be much more interesting than just saying "AI".<p>The comments here are really atrocious and ironically all seem LLM-generated.
This seems to be like kakasi for Japanese in that Japanese writing system does not separate words by spaces and one kanji can be read in multiple ways. As I understand the same is also true for cuneiform and this is an attempt to solve it.