Data Contamination in Neural Hieroglyphic Translation: A Reproducibility Study
Ammar Toutou, Abdelrahman Harb, Christine Basta · 2026
Read Data Contamination in Neural Hieroglyphic Translation: A Reproducibility Study by Ammar Toutou, Abdelrahman Harb, Christine Basta (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-07453.txt.
Description
Ammar Toutou, Abdelrahman Harb, and Christine Basta audit a striking translation result for Egyptian hieroglyphs and find data contamination behind it. A recent study reported a high BLEU score translating hieroglyphs into German with a fine-tuned model. The team's reproduction with the released model scored far lower, prompting a closer look at the data.
They found a share of test targets appearing identically in training data, and far more under loose overlap checks. Contaminated samples scored dramatically higher than clean ones across five configurations. Even document-level cleaning left many leaks, so target-level deduplication is required. A corrected clean test set and baselines give a realistic picture.
Catalog record
| CREATOR | Ammar Toutou, Abdelrahman Harb, Christine Basta |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 25 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.07453