Early-Exit and Instant Confidence Translation Quality Estimation
Vilém Zouhar, Maike Züfle, Beni Egressy, Julius Cheng, Mrinmaya Sachan, Jan Niehues · 2025
Read Early-Exit and Instant Confidence Translation Quality Estimation by Vilém Zouhar, Maike Züfle, Beni Egressy, Julius Cheng, Mrinmaya Sachan, Jan Niehues (2025) 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-2502-14429.txt.
Description
Vilém Zouhar and co-authors make translation quality estimation cheaper and self-aware in a 2025 paper. Quality estimation grades machine translations without a reference text, and it runs everywhere in large pipelines. Standard models are opaque and costly, which blocks their use at scale.
Their Instant Confidence COMET model adds an uncertainty score while matching older quality at a fraction of the cost. Early-Exit COMET goes further, producing scores and confidence at early layers of the network so easy cases can stop computing sooner. Paired with a bandit method for reranking translations, the tools cut compute in half with very little loss in quality. The work also shows how confidence scores can route doubtful translations to human reviewers.
Catalog record
| CREATOR | Vilém Zouhar, Maike Züfle, Beni Egressy, Julius Cheng, Mrinmaya Sachan, Jan Niehues |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 58 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-14429