Function-to-Style Guidance of LLMs for Code Translation

2025-07-15

AI Summary

Addressing the challenge of balancing semantic correctness and code readability in code translation, this paper proposes F2STrans—a novel paradigm that decouples functional learning (ensuring semantic equivalence) from style guidance (enhancing readability and consistency). Methodologically, it employs supervised learning to guarantee functional correctness and introduces contrastive learning with positive/negative style exemplars, augmented by a progressive prompting strategy to refine outputs. Contributions include: (1) the first fine-grained benchmark covering multiple programming languages, featuring high-quality human annotations, comprehensive unit tests, and state-of-the-art code; and (2) superior performance of the lightweight Qwen-1.5B model across 20 cross-language translation tasks—outperforming prompt-enhanced Qwen-32B and GPT-4 on average—demonstrating that structured learning enables compact models to achieve higher practical efficacy.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Large pretrained models with web data

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.

Problem

• Ensuring correctness and readability in LLM code translation
• Improving translation performance via function-to-style guidance
• Creating a benchmark for comprehensive translation evaluation

Innovation

• Function-to-style guiding paradigm for LLMs
• Two-stage learning: functional and style
• Novel benchmark with comprehensive evaluations