# The same study reports hallucinated output, weaker results on long and logogram-heavy lines, and test sentences drawn…

> The same study reports hallucinated output, weaker results on long and logogram-heavy lines, and test sentences drawn from the training corpora, so no…

Source: https://mysteriesdept.com/mysteries/unread-cuneiform/evidence/EVD0DEDXNXMGRAR/the-same-study-reports-hallucinated-output-weaker-results-on  
Publisher: Mysteries Department  
Licence: CC BY 4.0 (credit Mysteries Department and link the page)  

---

[Unread cuneiform: half a million tablets and the limits of Sumerian](https://mysteriesdept.com/mysteries/unread-cuneiform)Evidence

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

# The same study reports hallucinated output, weaker results on long and logogram-heavy lines, and test sentences drawn…

The same study reports hallucinated output, weaker results on long and logogram-heavy lines, and test sentences drawn from the training corpora, so no cross-genre or unpublished-text performance is known.

Supported0

## What it means

Key question[Can machines read the cuneiform tablets that no scholar has read?](https://mysteriesdept.com/mysteries/unread-cuneiform/questions/QST0DEDT2BJHV5Z/can-machines-read-the-cuneiform-tablets-that-no-scholar-has)

5 answers, none ahead yet

**Open** unchanged

Answer this evidence bears on[Machine translation scales to the backlog](https://mysteriesdept.com/mysteries/unread-cuneiform/answers/ANS0DEDW61E16Z8/machine-translation-scales-to-the-backlog)

Counts against2 for, 1 against

**Still possible** unchanged

## Next steps 5

1

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?](https://mysteriesdept.com/mysteries/unread-cuneiform/questions/QST0DEDT2BJHV5Z/can-machines-read-the-cuneiform-tablets-that-no-scholar-has)

**—** not judged

2

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?](https://mysteriesdept.com/mysteries/unread-cuneiform/questions/QST0DEDT2BJHV5Z/can-machines-read-the-cuneiform-tablets-that-no-scholar-has)

**—** not judged

3

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?](https://mysteriesdept.com/mysteries/unread-cuneiform/questions/QST0DEDT2BJHV5Z/can-machines-read-the-cuneiform-tablets-that-no-scholar-has)

**—** not judged

4

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?](https://mysteriesdept.com/mysteries/unread-cuneiform/questions/QST0DEDT2BJHV5Z/can-machines-read-the-cuneiform-tablets-that-no-scholar-has)

**—** not judged

Ranked by how likely each is to move the key question, for the time it takes[All 5 on the question](https://mysteriesdept.com/mysteries/unread-cuneiform/questions/QST0DEDT2BJHV5Z/can-machines-read-the-cuneiform-tablets-that-no-scholar-has#next)

## Sources 1

[Translating Akkadian to English with neural machine translation](https://mysteriesdept.com/sources/SRC0DEDGBY5JRWS/translating-akkadian-to-english-with-neural-machine)

Gutherz et al. 2023

SRC

## Cite it

```
Mysteries Department. (2026). The same study reports hallucinated output, weaker results on long and logogram-heavy lines, and test sentences drawn… [Evidence EVD0DEDXNXMGRAR]. https://mysteriesdept.com/r/EVD0DEDXNXMGRAR
```

Key

**EVD** 0DEDXNXMGRAR

Kind

Evidence

Tier

Supported

Version

1

Added by

[apollo](https://mysteriesdept.com/u/apollo)

The record

Every record is open data. Take this one as JSON, with its links and its history.

[JSON](https://mysteriesdept.com/api/v1/records/EVD0DEDXNXMGRAR)