Function-to-Style Guidance of LLMs for Code Translation

Longhui Zhang, Bin Wang, Jiahao Wang, Xiaofeng Zhao, Min Zhang, Hao Yang, Meishan Zhang, YU LI, Jing Li, Jun Yu, Min Zhang

Abstract: Large language models (LLMs) have made significant strides in code translation tasks. However, ensuring both the correctness and readability of translated code remains a challenge, limiting their effective adoption in real-world software development. In this work, we propose F2STrans, a function-to-style guiding paradigm designed to progressively improve the performance of LLMs in code translation. Our approach comprises two key stages: (1) Functional learning, which optimizes translation correctness using high-quality source-target code pairs mined from online programming platforms, and (2) Style learning, which improves translation readability by incorporating both positive and negative style examples. Additionally, we introduce a novel code translation benchmark that includes up-to-date source code, extensive test cases, and manually annotated ground-truth translations, enabling comprehensive functional and stylistic evaluations. Experiments on both our new benchmark and existing datasets demonstrate that our approach significantly improves code translation performance. Notably, our approach enables Qwen-1.5B to outperform prompt-enhanced Qwen-32B and GPT-4 on average across 20 diverse code translation scenarios.

Lay Summary:

Modern AI tools can already rewrite computer code from one language to another, yet the output usually forces developers to choose between “works correctly” or “easy to read”.
We created F2STrans, a two‑step training recipe that teaches an AI model to deliver both at once.
First, the model studies thousands of real programs and their trusted translations so it never changes what the code does.
Next, it sees examples of clean versus messy style, nudging it to write code that looks like something a seasoned programmer would proudly share.
By making automatic code translation reliable and pleasant to read, our work could speed up software maintenance, help companies modernize legacy projects, and let developers collaborate across programming languages more easily.

Primary Area: Deep Learning->Large Language Models

Keywords: Code Translation, Large Language Models

Submission Number: 4704