Unsupervised Translation of Programming Languages

2020/06/05 by Marie-Anne Lachaux, Lachaux, Marie-Anne, Baptiste Rozière +6 · 4 voices · 81 citations
Computer Science · #Abstract syntax tree #Artificial intelligence #COBOL #Computer science #Correctness #Machine translation #Natural Language Processing Techniques #Natural language processing #Open source #Parsing #Programming language #Python (programming language) #Software #Software Engineering Research #Source code #Topic Modeling #Unit testing #cs.CL #cs.PL

paper · pdf · doi:10.48550/arxiv.2006.03511

published in arXiv (Cornell University) (Cornell University)

openalex publication_date 2020/06/05 · arxiv created 2020/09/22 · arxiv updated 2020/09/23 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/05

Abstract

A transcompiler, also known as source-to-source translator, is a system that converts source code from a high-level programming language (such as C++ or Python) to another. Transcompilers are primarily used for interoperability, and to port codebases written in an obsolete or deprecated language (e.g. COBOL, Python 2) to a modern one. They typically rely on handcrafted rewrite rules, applied to the source code abstract syntax tree. Unfortunately, the resulting translations often lack readability, fail to respect the target language conventions, and require manual modifications in order to work properly. The overall translation process is timeconsuming and requires expertise in both the source and target languages, making code-translation projects expensive. Although neural models significantly outperform their rule-based counterparts in the context of natural language translation, their applications to transcompilation have been limited due to the scarcity of parallel data in this domain. In this paper, we propose to leverage recent approaches in unsupervised machine translation to train a fully unsupervised neural transcompiler. We train our model on source code from open source GitHub projects, and show that it can translate functions between C++, Java, and Python with high accuracy. Our method relies exclusively on monolingual source code, requires no expertise in the source or target languages, and can easily be generalized to other programming languages. We also build and release a test set composed of 852 parallel functions, along with unit tests to check the correctness of translations. We show that our model outperforms rule-based commercial baselines by a significant margin.

Citations

• Sequence to Sequence Learning with Neural Networks
• Learning Python Code Suggestion with a Sparse Pointer Network
• CodeBERT: A Pre-Trained Model for Programming and Natural Languages

Cited by

• Spaghetti Architect: A Contamination-Resistant, By-Construction-Labelled, Multi-Language Code Dataset Generator
• LLMs Corrupt Your Documents When You Delegate
• QiMeng: Fully Automated Hardware and Software Design for Processor Chip
• A Closer Look into Transformer-Based Code Intelligence Through Code Transformation: Challenges and Opportunities
• Neuron-Guided Interpretation of Code LLMs: Where, Why, and How?
• Beyond Language Boundaries: Uncovering Programming Language Families for Code Language Models
• SoK: Understanding (New) Security Issues Across AI4Code Use Cases
• TreeBERT: A Tree-Based Pre-Trained Model for Programming Language
• REMODEL-LLM: Transforming C code to Java using LLMs
• Bridging Code Graphs and Large Language Models for Better Code Understanding
• BabelCoder: Agentic Code Translation with Specification Alignment
• Beyond Code Pairs: Dialogue-Based Data Generation for LLM Code Translation
• Advancing Automated In-Isolation Validation in Repository-Level Code Translation
• Performance and interpretability analysis of code generation large language models
• Towards a Human-in-the-Loop Framework for Reliable Patch Evaluation Using an LLM-as-a-Judge
• An Ocean Model Ported by a Large Language Model: Experience and Lessons from FESOM2 (Fortran to C to C++/Kokkos)
• OMPILOT: Harnessing Transformer Models for Auto Parallelization to Shared Memory Computing Paradigms
• Automated Prompt Generation for Code Intelligence: An Empirical study and Experience in WeChat
• PortGPT: Towards Automated Backporting Using Large Language Models
• Unified Pre-training for Program Understanding and Generation
• Automated Snippet-Alignment Data Augmentation for Code Translation
• Project-Level C-to-Rust Translation via Synergistic Integration of Knowledge Graphs and Large Language Models
• TIT: A Tree-Structured Instruction Tuning Approach for LLM-Based Code Translation
• LLM Based Long Code Translation using Identifier Replacement
• RA-Gen: A Controllable Code Generation Framework Using ReAct for Multi-Agent Task Execution
• Adversarial Agent Collaboration for Correctness Improvements of C to Safe Rust Translation
• Can Emulating Semantic Translation Help LLMs with Code Translation? A Study Based on Pseudocode
• Evaluating Large Language Models Trained on Code
• Red Teaming Program Repair Agents: When Correct Patches can Hide Vulnerabilities
• Measuring Coding Challenge Competence With APPS
• Beyond Language Barriers: Multi-Agent Coordination for Multi-Language Code Generation
• MatchFixAgent: Language-Agnostic Autonomous Repository-Level Code Translation Validation and Repair
• RulER: Automated Rule-Based Semantic Error Localization and Repair for Code Translation
• Evaluating Large Language Models for Code Translation: Effects of Prompt Language and Prompt Design
• A New Benchmark for Evaluating Code Translation with Third-Party Libraries
• UniPar: A Unified LLM-Based Framework for Parallel and Accelerated Code Translation in HPC
• When Code Crosses Borders: A Security-Centric Study of LLM-based Code Translation
• Code Review Without Borders: Evaluating Synthetic vs. Real Data for Review Recommendation
• Neural Circuit Synthesis from Specification Patterns
• CoTexT: Multi-task Learning with Code-Text Transformer
• Model-Driven Quantum Code Generation Using Large Language Models and Retrieval-Augmented Generation
• FlowMalTrans: Unsupervised Binary Code Translation for Malware Detection Using Flow-Adapter Architecture
• Guidelines for Empirical Studies in Software Engineering involving Large Language Models
• TRACY: Benchmarking Execution Efficiency of LLM-Based Code Translation
• EvoC2Rust: A Skeleton-guided Framework for Project-Level C-to-Rust Translation
• Distilling Transformers for Neural Cross-Domain Search
• A Pilot Study on LLM-Based Agentic Translation from Android to iOS: Pitfalls and Insights
• On LLM-Assisted Generation of Smart Contracts from Business Processes
• On The Role of Pretrained Language Models in General-Purpose Text Embeddings: A Survey
• Compositional Generalization in Semantic Parsing: Pre-training vs. Specialized Architectures
• Automatic Generation of High-Performance RL Environments
• Automated Source Code Generation and Auto-completion Using Deep Learning: Comparing and Discussing Current Language-Model-Related Approaches
• QiMeng-MuPa: Mutual-Supervised Learning for Sequential-to-Parallel Code Translation
• ACT: Bridging the Gap in Code Translation through Synthetic Data Generation & Adaptive Training
• Learning Program Behavioral Models from Synthesized Input–Output Pairs
• Function-to-Style Guidance of LLMs for Code Translation
• LLM-Powered Quantum Code Transpilation
• Energy-Based Models for Code Generation under Compilability Constraints
• On the Reliability and Explainability of Language Models for Program Generation
• Mix-of-Language-Experts Architecture for Multilingual Programming
• Guaranteed Guess: A Language Modeling Approach for CISC-to-RISC Transpilation with Testing Guarantees
• Deployability-Centric Infrastructure-as-Code Generation: An LLM-based Iterative Framework
• Svar: A Tiny C++ Header Brings Unified Interface for Multiple programming Languages
• CETBench: A Novel Dataset constructed via Transformations over Programs for Benchmarking LLMs for Code-Equivalence Checking
• Text Data Augmentation for Deep Learning
• How Programming Concepts and Neurons Are Shared in Code Language Models
• Text Classification for Task-based Source Code Related Questions
• EffiBench-X: A Multi-Language Benchmark for Measuring Efficiency of LLM-Generated Code
• SafeTrans: LLM-assisted Transpilation from C to Rust
• Variational Prefix Tuning for Diverse and Accurate Code Summarization Using Pre-trained Language Models
• A Systematic Literature Review on Neural Code Translation
• QiMeng-Xpiler: Transcompiling Tensor Programs for Deep Learning Systems with a Neural-Symbolic Approach
• Frontier Coding Agents Use Metaprogramming to Adapt to Unfamiliar Programming Languages
• RepoRepair: Leveraging Code Documentation for Repository-Level Automated Program Repair
• A Catalog of Data Smells for Coding Tasks
• Report of the Workshop on Program Synthesis for Scientific Computing
• ReCodeAgent: A Multi-Agent Workflow for Language-agnostic Translation and Validation of Large-scale Repositories
• APIRAT: Integrating Multi-source API Knowledge for Enhanced Code Translation with LLMs
• OpenCodeInstruct: A Large-scale Instruction Tuning Dataset for Code LLMs
• Migrating Code At Scale With LLMs At Google
• Program Skeletons for Automated Program Translation

Discussions

• Unsupervised translation of programming languages [hn, 166 points, 44 comments]
• Unsupervised Translation of Programming Languages [pdf] [hn, 14 points, 1 comments]
• Unsupervised Translation of Programming Languages [hn, 3 points, 0 comments]
• Unsupervised Translation of Programming Languages [hn, 3 points, 0 comments]

Related

• Syntax and Domain Aware Model for Unsupervised Program Translation
• Leveraging Automated Unit Tests for Unsupervised Code Translation
• Exploring and Unleashing the Power of Large Language Models in Automated Code Translation
• User-Customizable Transpilation of Scripting Languages
• Program Translation via Code Distillation