Breaking the Programming Language Barrier: Multilingual Prompting to Empower Non-Native English Learners
Project Overview
The document explores the use of generative AI (GenAI) in education, particularly focusing on its potential to support non-native English speakers (NNES) in programming. It outlines the challenges NNES face, primarily stemming from the dominance of English in programming languages, which can hinder their learning experience. GenAI offers a solution by allowing these learners to interact with AI models in their native languages, such as Chinese, Portuguese, and Arabic, to address programming problems. The study indicates varying effectiveness across different language groups, with Portuguese and Chinese speakers showing more positive results compared to Arabic speakers. Additionally, it highlights a trade-off where NNES feel more expressive in their native languages but often achieve better accuracy when using English due to the syntax alignment of programming languages. Overall, the findings suggest that GenAI has the potential to democratize programming education, making it more inclusive and accessible for a diverse range of learners worldwide.
Key Applications
Prompt Problems - students write natural language prompts to generate code solutions.
Context: Programming education for non-native English speakers in various countries.
Implementation: Students were trained to use their native languages to write prompts for AI models to solve programming exercises. This was tested in classrooms across institutions in Portugal, New Zealand, and Saudi Arabia.
Outcomes: Students were able to solve programming problems effectively using native languages, with varying success rates across language groups. Engagement was reported as positive, especially for Portuguese and Chinese speakers.
Challenges: Arabic speakers faced greater challenges, often due to the lower performance of AI models with Arabic prompts and the complexity of translating programming concepts.
Implementation Barriers
Performance barrier
Generative AI models perform better with high-resource languages like English and Chinese, while lower-resource languages like Arabic face challenges due to limited training data. Additionally, while NNES find it easier to express ideas in their native languages, they achieve better outcomes when using English due to programming language syntax being primarily English-based.
Proposed Solutions: Future improvements in AI training data and model capabilities could enhance support for low-resource languages. Furthermore, developing pedagogies that incorporate multilingual support while acknowledging the need for technical precision in programming tasks could help bridge this gap.
Project Team
James Prather
Researcher
Brent N. Reeves
Researcher
Paul Denny
Researcher
Juho Leinonen
Researcher
Stephen MacNeil
Researcher
Andrew Luxton-Reilly
Researcher
João Orvalho
Researcher
Amin Alipour
Researcher
Ali Alfageeh
Researcher
Thezyrie Amarouche
Researcher
Bailey Kimmel
Researcher
Jared Wright
Researcher
Musa Blake
Researcher
Gweneth Barbre
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: James Prather, Brent N. Reeves, Paul Denny, Juho Leinonen, Stephen MacNeil, Andrew Luxton-Reilly, João Orvalho, Amin Alipour, Ali Alfageeh, Thezyrie Amarouche, Bailey Kimmel, Jared Wright, Musa Blake, Gweneth Barbre
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai