BanglaSTEM: A Parallel Corpus for Technical Domain Bangla-English Translation

2025-11-05

AI Summary

Existing Bangla–English translation systems exhibit low accuracy in STEM-domain terminology, causing semantic distortion in technical queries and hindering Bangla-speaking users’ ability to effectively leverage English large language models for domain-specific problem solving. Method: We construct the first high-quality, STEM-focused (computer science, mathematics, physics, etc.) Bangla–English parallel corpus comprising 5,000 sentence pairs. Our curation employs a human-in-the-loop pipeline integrating large language model–assisted generation with rigorous human verification to ensure terminological precision and semantic fidelity. A dedicated translation model is trained on the T5 architecture using this corpus. Contribution/Results: On code generation and mathematical problem-solving tasks, our model achieves a +12.3 BLEU improvement over strong baselines and reduces critical terminology error rate by 67%. Both the dataset and model are publicly released, establishing a reproducible benchmark and practical resource for low-resource language technical translation.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: (Computational) Cognitive Architectures

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Large language models work well for technical problem solving in English but perform poorly when the same questions are asked in Bangla. A simple solution would be to translate Bangla questions into English first and then use these models. However, existing Bangla-English translation systems struggle with technical terms. They often mistranslate specialized vocabulary, which changes the meaning of the problem and leads to wrong answers. We present BanglaSTEM, a dataset of 5,000 carefully selected Bangla-English sentence pairs from STEM fields including computer science, mathematics, physics, chemistry, and biology. We generated over 12,000 translations using language models and then used human evaluators to select the highest quality pairs that preserve technical terminology correctly. We train a T5-based translation model on BanglaSTEM and test it on two tasks: generating code and solving math problems. Our results show significant improvements in translation accuracy for technical content, making it easier for Bangla speakers to use English-focused language models effectively. Both the BanglaSTEM dataset and the trained translation model are publicly released at https://huggingface.co/reyazul/BanglaSTEM-T5.

Problem

• Existing Bangla-English translation systems struggle with technical terminology
• Technical term mistranslation alters problem meaning and causes incorrect answers
• BanglaSTEM addresses specialized vocabulary translation in STEM fields

Innovation

• Created parallel corpus for technical Bangla-English translation
• Trained T5 model on curated STEM domain dataset
• Improved translation accuracy for specialized technical terminology