TabletCraft: Bridging a 4,000-Year Cultural Gap with Bidirectional Akkadian NMT and Cuneiform Rendering

What it is about

TabletCraft is an open-source toolkit that works in both directions between Akkadian and English, and renders the result as a cuneiform tablet image. A ByT5-base model (581M parameters) is fine-tuned on 116K bidirectional pairs made from 58K aligned sentences, mostly Neo-Assyrian royal inscriptions, with task prefixes for each direction. A lookup table of 14,240 transliteration-to-cuneiform mappings turns the output into Unicode signs, and an SVG or PNG renderer draws the tablet. On the 2,812-sample Akkademia validation split, the model scores 49.1 BLEU for Akkadian to English and 48.5 BLEU (55.6 chrF++) for English to Akkadian. The converter covers 95.3% of tokens on 1,000 test transliterations. The authors state that the English-to-Akkadian score is a lower bound, because each source has one reference and Akkadian allows several spellings of the same word.

Ideas

• I1. Coverage as an output of every rule: for each text, report the share of tokens that a lookup table covered, and pass the uncovered tokens on marked, not silently. The Akkadian treebank converter in treebanks/akk should print this coverage with every run.
• I2. Multi-reference test for generation: for each lemma, list the alternative spellings (logogram or syllabic, LUGAL or MAN), and score the rule-based generator against all of them. Report the single-reference and multi-reference scores side by side.
• I3. Round trip from the world: the English-to-Akkadian direction is a generation step. Its Mova counterpart is to render frames from prag and world back into English and score the result against the source sentence, deterministically. This is the same residual check as the round-trip autoencoder.

For Mova

I1 and I2 go into treebanks/akk, which has no Akkadian parser or generator yet; the test set is part of that. I3 belongs to world and prag (English) as a generation check, reported beside the parser's own round trip.

Topics: Akkadian and cuneiform NLP · Machine translation

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.c3nlp-1.10

@inproceedings{wang2026tabletcraft,
  title = {{TabletCraft: Bridging a 4,000-Year Cultural Gap with Bidirectional Akkadian NMT and Cuneiform Rendering}},
  author = {Wang, Zhaohui Geoffrey},
  booktitle = {Proceedings of the 4th Workshop on Cross-Cultural Considerations in NLP (C3NLP 2026)},
  year = {2026},
  doi = {10.18653/v1/2026.c3nlp-1.10},
  url = {https://doi.org/10.18653/v1/2026.c3nlp-1.10}
}

References

20 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Emily M. Bender, Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell. 2021. On the dangers of stochastic parrots: Can language models be too big? In Proceedings of FAccT, pages 610–623. find
2. Steven Bird. 2020. Decolonising speech and language technology. In Proceedings of the 28th International Conference on Computational Linguistics, pages 3504–3519. find
3. CDLI. 2024. CDLI: Cuneiform digital library initiative. find
4. Deep Past Initiative. 2026. Deep past challenge: Translate Akkadian to English. Kaggle Competition. find
5. Hadi Farsi and 1 others. 2025. ParsiPy: NLP toolkit for historical Persian texts in Python. In Proceedings of the Second Workshop on Ancient Language Processing. Association for Computational Linguistics. find
6. Shai Gordin, Adam Anderson, and 1 others. 2025. Eva-Cun 2025 shared task: Lemmatization and token prediction in Akkadian and Sumerian using LLMs. In Proceedings of the Second Workshop on Ancient Language Processing. Association for Computational Linguistics. find
7. Shai Gordin, Gai Gutherz, Ariel Elazary, Avital Romach, Enrique Jiménez, Jonathan Berant, and Luis Sáenz. 2020. Reading Akkadian cuneiform using natural language processing. PLoS ONE, 15(10):e0240511. find
8. Gai Gutherz, Shai Gordin, Luis Sáenz, Omer Levy, and Jonathan Berant. 2023. Translating Akkadian to English with neural machine translation. PNAS Nexus, 2(5):pgad096. find
9. Louise Rayne, Jennie Bradbury, David Mattingly, Graham Philip, Robert Bewley, and Andrew Wilson. 2017. From above and on the ground: Geospatial methods for recording endangered archaeology in the Middle East and North Africa. Geosciences, 7(4):100. find
10. Frederick Riemenschneider. 2025. Beyond base predictors: Using LLMs to resolve ambiguities in Akkadian lemmatization. In Proceedings of the Second Workshop on Ancient Language Processing. Association for Computational Linguistics. find our card
11. Rico Sennrich, Barry Haddow, and Alexandra Birch. 2016. Improving neural machine translation models with monolingual data. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 86–96. find
12. Thea Sommerschield, Yannis Assael, John Pavlopoulos, Vanessa Stefanak, Andrew Senior, Chris Dyer, John Bodel, Jonathan Prag, Ion Androutsopoulos, and Nando de Freitas. 2023. Machine learning for ancient languages: A survey. Computational Linguistics, 49(3):703–747. find
13. Melissa Terras. 2012. Digitisation and digital resources in the humanities. In Claire Warwick, Melissa Terras, and Julianne Nyhan, editors, Digital Humanities in Practice, pages 47–70. Facet Publishing. find
14. Klaas R. Veenhof and Jesper Eidem. 2008. Mesopotamia: The Old Assyrian Period. Academic Press Fribourg. find
15. C. B. F. Walker. 1987. Cuneiform. University of California Press. find
16. Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, and Colin Raffel. 2022. ByT5: Towards a token-free future with pre-trained byte-to-byte models. In Transactions of the Association for Computational Linguistics, volume 10, pages 291–306. find
17. Murat Yavasan and Shai Gordin. 2025. From clay to code: Transforming Hittite texts for machine learning. In Proceedings of the Second Workshop on Ancient Language Processing. Association for Computational Linguistics. find
18. John Huehnergard. 2011. A Grammar of Akkadian, 3rd edition. Eisenbrauns. find
19. Pengxiu Lu, Yonglong Huang, Jing Xu, Minxuan Feng, and Chao Xu. 2025. Lemmatization of cuneiform languages using the ByT5 model. In Proceedings of the Second Workshop on Ancient Language Processing, pages 197–205. Association for Computational Linguistics. find
20. ORACC Project. 2024. ORACC: Open richly annotated cuneiform corpus. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/coref-2026-wang-tabletcraft-bridging-4-000-year/