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

Yuhan Wu, Huan Zhang, Wei Cheng, Chen Shen, Jingyue Yang, Wei Hu · 2026

Read Improving Code Translation with Syntax-Guided and Semantic-aware Preference Optimization by Yuhan Wu, Huan Zhang, Wei Cheng, Chen Shen, Jingyue Yang, Wei Hu (2026) 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-2605-13229.txt.

Description

Wu and colleagues present CTO, a method for translating code between languages with fewer errors. Language models often break syntax or drift from the source meaning, and sparse tests give weak feedback on meaning. CTO trains a cross language model to judge whether source and translated code do the same thing. It pairs that meaning signal with compiler feedback inside preference training, beating baselines across C++, Java and Python.

Catalog record

CREATORYuhan Wu, Huan Zhang, Wei Cheng, Chen Shen, Jingyue Yang, Wei Hu
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE36817
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.13229

Browse the full library · About the Tower of Babel