boubacar-toulba/Hassaniya-Translator

A neural machine translation system for Hassaniya Arabic dialect spoken in Mauritania to English, built with PyTorch and Seq2Seq architecture with attention mechanism. # Hassaniya-Translator # 🌙 Hassaniya-Bridge: Neural Machine Translation for Hassaniya Arabic A neural machine translation system for Hassaniya Arabic (a Maghrebi Arabic dialect spoken in Mauritania and Western Sahara) to English, built with PyTorch and Seq2Seq architecture with attention mechanism. ## 🌟 Project Overview **Hassaniya-Bridge** aims to bridge the language gap for one of the world's under-resourced languages. Hassaniya Arabic, spoken by over 3 million people across West Africa, has minimal digital presence and virtually no machine translation support. This project represents a step toward digital inclusion and language preservation. ### Why This Matters - 🗣️ **Language Preservation**: Hassaniya is an oral tradition language with limited written resources - 🌍 **Digital Inclusion**: Enables Hassaniya speakers to access global information - 📚 **Research Foundation**: First open-source NMT system for this dialect - 🤝 **Cultural Bridge**: Connects Hassaniya-speaking communities with the wider world ## ✨ Features - ⚡ **Seq2Seq with Attention**: LSTM-based encoder-decoder with Bahdanau attention - 📊 **Custom Vocabulary Builder**: Handles Arabic script and dialect-specific terms - 🎯 **Teacher Forcing**: Accelerated training with configurable forcing ratio - 💾 **Model Checkpointing**: Automatic saving of best models - 📈 **Training Visualization**: Real-time loss tracking and metrics - 🔍 **Inference Pipeline**: Easy-to-use translation interface ## 🚀 Quick Start ### Prerequisites ```bash python >= 3.8 torch >= 2.0.0 numpy >= 1.21.0 matplotlib >= 3.5.0 ``` ### Installation ```bash # Clone the repository git clone github.com cd hassaniya-bridge # Install dependencies pip install -r requirements.txt ``` ### Basic Usage ```python from model import Seq2Seq, Encoder, Decoder, Attention from utils import translate_sentence # Load trained model model = torch.load('best_model.pt') # Translate a sentence hassa …