Improving Code Translation with Syntax-Guided and Semantic-aware Preference Optimization

2026-05-13 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Expert, quick

Summary

A novel approach called CTO (Code Translation Optimization) has been developed to enhance large language model (LLM) performance in code translation, specifically addressing challenges in syntactic correctness and semantic consistency. Traditional preference-based learning methods are often limited by unreliable semantic rewards, which typically stem from sparse test cases or overly restrictive reference translations. CTO introduces a robust semantic reward system derived directly from the source code. It employs contrastive learning to train a cross-lingual semantic model, enabling direct assessment of functional equivalence between source and translated code. This semantic signal is then integrated with compiler-based syntactic feedback within a direct preference optimization framework, treating code translation as a multi-objective optimization problem. Experiments across C++, Java, and Python translations show CTO significantly outperforms current baselines and other preference optimization strategies.

Key takeaway

For research scientists developing code translation models, CTO offers a critical advancement by demonstrating how to achieve superior syntactic correctness and semantic consistency. You should consider integrating source-derived semantic rewards and compiler-based syntactic feedback into your preference optimization frameworks to overcome limitations of traditional test-case-dependent methods, potentially leading to more robust and functionally equivalent code translations.

Key insights

CTO improves code translation by unifying syntax-guided and source-derived semantic feedback within a preference optimization framework.

Principles

• Semantic rewards should derive from source code.
• Functional equivalence is key for code translation.
• Multi-objective optimization enhances code translation.

Method

CTO trains a cross-lingual semantic model via contrastive learning to assess functional equivalence, then unifies this semantic signal with compiler-based syntactic feedback using direct preference optimization.

In practice

• Use contrastive learning for cross-lingual semantic models.
• Integrate compiler feedback for syntactic correctness.
• Apply multi-objective optimization to code tasks.

Topics

• Code Translation
• Preference Optimization
• Semantic Equivalence
• Syntax-Guided Optimization
• Cross-lingual Models

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• Improving Code Translation with Syntax-Guided and Semantic-aware Preference Optimization — CTO unifies syntax-guided and semantic-aware preference optimization for robust, accurate code translation.
• Improving Code Translation with Syntax-Guided and Semantic-aware Preference Optimization — CTO improves code translation by unifying syntax-guided and semantic-aware preference optimization.
• Zero-Shot to Full-Resource: Cross-lingual Transfer Strategies for Aspect-Based Sentiment Analysis — Cross-lingual transfer strategies for ABSA vary in effectiveness based on model architecture and resource availability.
• Translating the Untranslatable: An Operationalizable Ontology for Untranslatability — A new ontology and dataset enable strategy-informed machine translation for untranslatable linguistic phenomena.
• A Dataset for Probing Translationese Preferences in English-to-Swedish Translation — Language models often prefer literal "translationese" over idiomatic phrasing, especially when exposed to the source language.
• Macro: Enhancing Multilingual Counterfactual Explanations through Alignment-as-Preference Optimization — Preference alignment, specifically DPO, effectively balances validity and minimality in multilingual counterfactual explanation generation for LLMs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.