Multilingual Natural Language Interfaces for Code Generation: Building Inclusive AI Tools for Global Developers

Abstract

The dominance of English in artificial intelligence (AI) tools for code generation has created an accessibility gap for non-English-speaking developers worldwide. As global programming communities continue to expand, the need for inclusive and linguistically diverse developer tools becomes increasingly critical. This paper introduces a lightweight, multilingual Natural Language Interface (NLI) framework designed to translate natural language prompts in various languages into executable code. Leveraging state-of-the-art multilingual large language models (LLMs) such as mT5 and mBART, the proposed system is fine-tuned using a novel dataset of parallel programming prompts in English, Spanish, French, Hindi, Yoruba, Arabic, and Mandarin. Both automatic (BLEU, CodeBLEU, execution accuracy) and human-centered (readability, syntactic and logical correctness) evaluations demonstrate that while English inputs yield the highest performance, several other languages show promising results with moderate adaptations.

Additionally, real-world case studies from educational platforms, coding bootcamps, and regional hackathons reveal the practical utility and social impact of deploying such inclusive tools. The findings highlight performance disparities across languages, especially in low-resource linguistic settings, and suggest pathways for improving fairness and accuracy through language-aware tokenization, script normalization, and culturally sensitive prompt engineering. By democratizing access to AI-assisted code generation, this research contributes to bridging the digital divide in software development and fosters a more inclusive global programming ecosystem.