The Art of Asking: Multilingual Prompt Optimization for Synthetic Data
This addresses the challenge of building more robust and culturally grounded multilingual LLMs, representing an incremental improvement over existing translation-based methods.
The paper tackled the problem of multilingual synthetic data generation being bottlenecked by translation-based prompts, which limit model generalization, by introducing a lightweight prompt-space optimization framework that improved downstream performance across 12 languages, achieving gains such as +4.7% on Global-MMLU accuracy and +35.3% wins in preferences on mArenaHard.
Synthetic data has become a cornerstone for scaling large language models, yet its multilingual use remains bottlenecked by translation-based prompts. This strategy inherits English-centric framing and style and neglects cultural dimensions, ultimately constraining model generalization. We argue that the overlooked prompt space-the very inputs that define training distributions-offers a more powerful lever for improving multilingual performance. We introduce a lightweight framework for prompt-space optimization, where translated prompts are systematically transformed for Naturalness, Cultural Adaptation, and Difficulty Enhancement. Using an off-the-shelf multilingual LLM, we apply these transformations to prompts for 12 languages spanning 7 families. Under identical data conditions, our approaches achieve substantial and consistent downstream improvements over the translation-only baseline: +4.7% on Global-MMLU accuracy, +2.4% on Flores XCometXL and +35.3% wins in preferences on mArenaHard. We establish prompt-space optimization as a simple yet powerful paradigm for building multilingual LLMs that are more robust, culturally grounded, and globally capable.