Evidence · Unread cuneiform: half a million tablets and the limits of Sumerian

Both restoration studies (recurrent networks 2020, masked language models 2021) present their output as suggestions…

Both restoration studies (recurrent networks 2020, masked language models 2021) present their output as suggestions for editors and evaluate them by masked held-out text and expert judgement, not by adoption in editions.

Suggestive0

What it means

Key questionCan machines read the cuneiform tablets that no scholar has read?

5 answers, none ahead yet

Openunchanged

Answer this evidence bears onMachines as assistants, reading stays human

Supports3 for, none against

Still possibleunchanged

Next steps 5

Genre-transfer test: train a transliteration-to-English model on the Akkademia ORACC splits (Neo-Assyrian royal inscriptions), evaluate unchanged on ORACC Old Babylonian letters and on SumTablets-linked Ur III texts with English, and report BLEU/chrF against the in-genre held-out score.

~16 h

On: Can machines read the cuneiform tablets that no scholar has read?

—not judged

Run the genre-transfer test of Akkadian NMT on Old Babylonian letters and Ur III texts.

~16 h

On: Can machines read the cuneiform tablets that no scholar has read?

—not judged

Glyph-to-transliteration baseline on the stored SumTablets test split (sha256 65dfe606…): train on the public train split, report sign-reading accuracy by period and genre, and compare with Akkademia's 96.7 per cent on RINAP to see how much reading Sumerian signs costs over Akkadian.

~10 h

On: Can machines read the cuneiform tablets that no scholar has read?

—not judged

Measure the backlog: from the CDLI bulk dump (cdli_cat.csv plus cdliatf_unblocked.atf) tabulate catalogued artifacts by period and genre with the share that has ATF, and reconcile the Ur III row with BDTNS's 105,187 / 75,818 figures.

~6 h

On: Can machines read the cuneiform tablets that no scholar has read?

—not judged

Ranked by how likely each is to move the key question, for the time it takesAll 5 on the question

Sources 2

Filling the Gaps in Ancient Akkadian Texts: A Masked Language Modelling Approach

Lazar et al. 2021

SRC

Restoration of fragmentary Babylonian texts using recurrent neural networks

Fetaya et al. 2020

SRC

Cite it

Mysteries Department. (2026). Both restoration studies (recurrent networks 2020, masked language models 2021) present their output as suggestions… [Evidence EVD0DEDY3FKDT9Y]. https://mysteriesdept.com/r/EVD0DEDY3FKDT9Y

@misc{mysteriesdept_evd0dedy3fkdt9y,
  title     = {Both restoration studies (recurrent networks 2020, masked language models 2021) present their output as suggestions…},
  author    = {{Mysteries Department}},
  year      = {2026},
  note      = {Evidence, catalog key EVD0DEDY3FKDT9Y},
  url       = {https://mysteriesdept.com/r/EVD0DEDY3FKDT9Y}
}