Beyond English: The Impact of Prompt Translation Strategies across Languages and Tasks in Multilingual LLMs
Itai Mondshine, Tzuf Paz-Argaman, Reut Tsarfaty · 2025
Read Beyond English: The Impact of Prompt Translation Strategies across Languages and Tasks in Multilingual LLMs by Itai Mondshine, Tzuf Paz-Argaman, Reut Tsarfaty (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-2502-09331.txt.
Description
Mondshine, Paz-Argaman, and Tsarfaty ask when translating a prompt into English helps a multilingual language model. Many users pre-translate prompts because models still favor English. Others translate only parts of the prompt, with little evidence to guide the choice.
The team split prompts into four parts: instruction, context, examples, and output format. They tested every mix across 35 languages and four tasks, from question answering to summarization. Results turned on how close each language sits to English, how good the translation was, and how much training data the language had. The authors close with practical rules for picking a strategy in each setting.
Catalog record
| CREATOR | Itai Mondshine, Tzuf Paz-Argaman, Reut Tsarfaty |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 85 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-09331