Building large-scale English–Romanian literary translation resources with open models
Mihai Nadaş, Laura Dioşan, Andreea Tomescu, Andrei Pişcoran · 2026
Read Building large-scale English–Romanian literary translation resources with open models by Mihai Nadaş, Laura Dioşan, Andreea Tomescu, Andrei Pişcoran (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/PMC13318819.txt.
Description
Mihai Nadaş and colleagues present a 2026 framework for literary translation from English into Romanian using open models. Romanian is a low-resource language for creative text. Their process builds data, fine-tunes a model, and tests results. It starts from a large pool of synthetic English fables.
A strong large model made 15,000 high-quality Romanian sample translations. A 12-billion-parameter open model was then fine-tuned in two stages. First-stage tuning taught fable style, and adapter shrinking eased rollout. Judging used a five-part rubric scoring accuracy, fluency, flow, style, and cultural fit. The tuned model nearly matched top commercial systems on fluency and adequacy. The team put the model, two datasets, and all scripts online for anyone to use.
Catalog record
| CREATOR | Mihai Nadaş, Laura Dioşan, Andreea Tomescu, Andrei Pişcoran |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 79719 |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |