Multilingual Machine Translation with Open Large Language Models at Practical Scale: An Empirical Study

Menglong Cui, Pengzhi Gao, Wei Liu, Jian Luan, Bin Wang · 2025

Read Multilingual Machine Translation with Open Large Language Models at Practical Scale: An Empirical Study by Menglong Cui, Pengzhi Gao, Wei Liu, Jian Luan, Bin Wang (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-2502-02481.txt.

Description

Menglong Cui and co-authors test open language models as translators in a 2025 study. Small open models, under ten billion parameters, have grown fast at handling many languages. The team ran six such models through wide translation tests. One, Gemma2-9B, stood out with strong scores across language pairs.

They then kept training with a data mix that puts parallel text first and single-language text second. The result, GemmaX2-28, covers 28 languages. It beat the top specialist models in their tests, such as TowerInstruct and XALMA. It also stood close to Google Translate and GPT-4-turbo, two closed giants. The study argues open models at handy scale can now anchor real translation work.

Catalog record

CREATORMenglong Cui, Pengzhi Gao, Wei Liu, Jian Luan, Bin Wang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE72 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-02481

Browse the full library · About the Tower of Babel