TabletCraft lets users write and read 4,000-year-old cuneiform

Researcher Zhaohui Wang has developed TabletCraft, a groundbreaking open-source toolkit that bridges a 4,000-year linguistic gap by enabling bidirectional interaction with ancient cuneiform script. The system addresses a critical limitation in existing natural language processing tools, which have only supported one-way translation from Akkadian to English. TabletCraft's architecture combines three core components: a ByT5-based neural machine translation (NMT) model trained on 116,000 bidirectional samples, a cuneiform sign converter with 14,240 mappings achieving 95.3% coverage, and a visual tablet renderer that generates realistic clay tablet representations.

On a held-out validation set of 2,812 Akkadian samples (Akkademia split), the system achieved 49.1 BLEU for Akkadian-to-English translation and 48.5 BLEU for the reverse direction—marking the first published quantitative results for English-to-Akkadian translation. Packaged as a pip-installable toolkit with both CLI and web demo interfaces, TabletCraft democratizes access to Mesopotamian culture, allowing non-specialists to actively participate in ancient writing rather than merely consuming translated content. The system was accepted to C3NLP @ ACL 2026 and represents a significant advancement in computational linguistics and cultural heritage preservation.