Hallucinations of Large Language Models in Cross-Language Code Translation: Phenomena and Mitigation

Key points

• Code translation converts source code from one programming language to another while preserving its functionality.1
• This paper presents a systematic empirical study of LLM hallucinations in code translation.1
• Guided by the taxonomy, we further propose HaluCodeTran, a mitigation framework that combines BM25-based retrieval of relevant translation examples with taxonomy-aware reflective refinement.1
• The study provides an execution-tested characterization of code-translation hallucinations and a practical baseline for building more reliable LLM-based code migration tools.1

Sources (1)

• [1]Hallucinations of Large Language Models in Cross-Language Code Translation: Phenomena and Mitigation
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 08:50 AM

  Although Large Language Models (LLMs) have improved code translation, their outputs still contain hallucinations: generated code that conflicts with the source program, violates target-language facts, or introduces contextually inconsistent dependencies.

  Code translation converts source code from one programming language to another while preserving its functionality.

Extractive summary: sentences quoted from the sources.