Large Language Models for Multilingual Code Intelligence: A Survey

TL;DR

This survey reviews the use of large language models in multilingual code intelligence, emphasizing challenges in less-resourced languages and the importance of semantic preservation across languages.

Contribution

It provides a comprehensive overview of methods, benchmarks, and challenges in multilingual code generation and translation using large language models.

Findings

Current models perform well on high-resource languages like Python.

Multilingual code translation aims to preserve semantics across languages.

Challenges include bias toward high-resource languages and ensuring trustworthy generalization.

Abstract

Large language models have transformed AI-assisted software engineering, but current research remains biased toward high-resource languages such as Python, with weaker performance in languages like Rust and OCaml. Since real-world systems are inherently polyglot, robust multilingual code intelligence is crucial. This survey focuses on two key tasks: multilingual code generation from shared natural-language requirements, and multilingual code translation that preserves semantics across languages. It reviews representative methods, benchmarks, and evaluation metrics, and highlights challenges and opportunities for trustworthy cross-language generalization.