Skip to main content Skip to navigation

Empowering Bengali Education with AI: Solving Bengali Math Word Problems through Transformer Models

Project Overview

The document explores the application of generative AI in education, specifically focusing on a tool designed to solve Bengali math word problems through advanced transformer models. It introduces the 'PatiGonit' dataset, which comprises 10,000 math problems in Bengali, and assesses the effectiveness of several transformer-based models, including mT5, BanglaT5, and mBART50. The findings indicate that the mT5 model achieved the highest accuracy of 97.30%, demonstrating its potential to significantly improve math education for Bengali-speaking students. This research not only aims to enhance the learning experience in mathematics for this demographic but also contributes to the broader field of Bengali natural language processing (NLP), illustrating the transformative impact of generative AI technologies in educational contexts.

Key Applications

PatiGonit dataset and transformer models for solving Bengali math word problems

Context: Targeting Bengali-speaking elementary students to improve math education through AI

Implementation: Development of transformer models (mT5, BanglaT5, mBART50) fine-tuned on the PatiGonit dataset

Outcomes: Achieved high accuracy in solving Bengali math word problems, significant improvement in educational tools for low-resource languages.

Challenges: The model faced difficulties in translating and culturally adapting math problems due to linguistic variations and complexities.

Implementation Barriers

Cultural and Linguistic

Translating English math word problems into Bengali posed challenges due to cultural context differences.

Proposed Solutions: Careful linguistic adjustments and dataset curation to ensure cultural relevance.

Dataset Limitations

The dataset predominantly featured basic arithmetic problems, limiting the models' capacity to handle more complex multi-step problems. Future dataset expansion is needed to include a broader range of mathematical challenges.

Proposed Solutions: Expand the dataset to incorporate a wider variety of mathematical problems.

Project Team

Jalisha Jashim Era

Researcher

Bidyarthi Paul

Researcher

Tahmid Sattar Aothoi

Researcher

Mirazur Rahman Zim

Researcher

Faisal Muhammad Shah

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Jalisha Jashim Era, Bidyarthi Paul, Tahmid Sattar Aothoi, Mirazur Rahman Zim, Faisal Muhammad Shah

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

Let us know you agree to cookies