Algorithm-Based Pipeline for Reliable and Intent-Preserving Code Translation with LLMs

Researchers Shahriar Rumi Dipto, Saikat Mondal, and Chanchal K. Roy developed an algorithm-based pipeline for code translation with LLMs. It introduces a language-neutral intermediate specification to capture program intent before generation. Testing on Python-Java translation with five LLMs, it increased micro-average accuracy from 67.7% to 78.5%, eliminated lexical errors by 100%, and reduced structural issues by 61.1%. This enables more reliable multilingual programming assistants.