Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Systematic Literature Review on Neural Code Translation

Xiang Chen, Jiacheng Xue, Xiaofei Xie, Caokai Liang, Xiaolin Ju

TL;DR

This systematic literature review analyzes 57 neural code translation studies from 2020 to 2025 to map the field across task characteristics, data preprocessing, code modeling, model construction, post-processing, evaluation subjects, and evaluation metrics. It finds that bidirectional translations between statically typed languages at function-level granularity dominate, with data cleaning and deduplication as key preprocessing steps, and plain-text modeling being most common alongside growing graph-based and IR-based approaches. Model construction trends favor fine-tuning and prompt engineering, while post-processing and robust evaluation (static and dynamic metrics) are increasingly emphasized to ensure correctness and maintainability. The study highlights open directions including leveraging retrieval-augmented generation and multi-agent LLM systems, improving translations for low-resource languages, securing code translation pipelines, and incorporating repository-level context and repair/refinement processes for practical, large-scale software migration.

Abstract

Code translation aims to convert code from one programming language to another automatically. It is motivated by the need for multi-language software development and legacy system migration. In recent years, neural code translation has gained significant attention, driven by rapid advancements in deep learning and large language models. Researchers have proposed various techniques to improve neural code translation quality. However, to the best of our knowledge, no comprehensive systematic literature review has been conducted to summarize the key techniques and challenges in this field. To fill this research gap, we collected 57 primary studies covering the period 2020~2025 on neural code translation. These studies are analyzed from seven key perspectives: task characteristics, data preprocessing, code modeling, model construction, post-processing, evaluation subjects, and evaluation metrics. Our analysis reveals current research trends, identifies unresolved challenges, and shows potential directions for future work. These findings can provide valuable insights for both researchers and practitioners in the field of neural code translation.

A Systematic Literature Review on Neural Code Translation

TL;DR

This systematic literature review analyzes 57 neural code translation studies from 2020 to 2025 to map the field across task characteristics, data preprocessing, code modeling, model construction, post-processing, evaluation subjects, and evaluation metrics. It finds that bidirectional translations between statically typed languages at function-level granularity dominate, with data cleaning and deduplication as key preprocessing steps, and plain-text modeling being most common alongside growing graph-based and IR-based approaches. Model construction trends favor fine-tuning and prompt engineering, while post-processing and robust evaluation (static and dynamic metrics) are increasingly emphasized to ensure correctness and maintainability. The study highlights open directions including leveraging retrieval-augmented generation and multi-agent LLM systems, improving translations for low-resource languages, securing code translation pipelines, and incorporating repository-level context and repair/refinement processes for practical, large-scale software migration.

Abstract

Code translation aims to convert code from one programming language to another automatically. It is motivated by the need for multi-language software development and legacy system migration. In recent years, neural code translation has gained significant attention, driven by rapid advancements in deep learning and large language models. Researchers have proposed various techniques to improve neural code translation quality. However, to the best of our knowledge, no comprehensive systematic literature review has been conducted to summarize the key techniques and challenges in this field. To fill this research gap, we collected 57 primary studies covering the period 2020~2025 on neural code translation. These studies are analyzed from seven key perspectives: task characteristics, data preprocessing, code modeling, model construction, post-processing, evaluation subjects, and evaluation metrics. Our analysis reveals current research trends, identifies unresolved challenges, and shows potential directions for future work. These findings can provide valuable insights for both researchers and practitioners in the field of neural code translation.
Paper Structure (23 sections, 9 figures, 3 tables)

This paper contains 23 sections, 9 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Code Translation
  4. Neural Machine Translation
  5. Review Methodology
  6. Research Questions
  7. Search Strategy
  8. Study Selection
  9. Data Extraction
  10. Data Synthesis
  11. Result Analysis
  12. RQ1: Task Characteristics
  13. Programming Language Pairs
  14. Code Granularity
  15. RQ2: Data Preprocessing
  16. ...and 8 more sections

Figures (9)

  • Figure 1: Illustrative example of code translation from Java to Python
  • Figure 2: Overview of the primary study search and collection process
  • Figure 3: Publication trends and major venues, (a) publication number over different years, (b) word cloud of major venues
  • Figure 4: Task characteristic summary of neural code translation in terms of code granularity
  • Figure 5: Data preprocessing summary of neural code translation
  • ...and 4 more figures