Adeyemi0/Yoruba-English-Translation

# 🌍 Yoruba ⇄ English Bidirectional Translation Model A production-ready neural machine translation system for Yoruba-English bidirectional translation, built by fine-tuning AfriTeVa V2 on 101,906 translation pairs. **Live Demo:** Try it on Hugging Face Spaces --- ## 📋 Table of Contents - Overview - Key Features - Model Performance - Architecture - Installation - Quick Start - API Reference - Web Interface - Training Details - Dataset - Technical Implementation - Deployment - Known Limitations - Future Improvements - Contributing - Citation - License - Connect --- ## 🎯 Overview This project implements neural machine translation system for Yoruba-English language pairs. The model supports: - **Bidirectional Translation**: Seamless English → Yoruba and Yoruba → English translation - **Automatic Language Detection**: Intelligently detects input language using character patterns and linguistic features - **Multiple Translation Variants**: Generates 4 different translations with quality scoring to give users options - **Smart Text Chunking**: Handles long texts through intelligent sentence-based segmentation - **Production-Ready API**: RESTful Flask API with CORS support and comprehensive error handling - **Modern Web Interface**: Beautiful, responsive UI with real-time translation and diacritical mark support Built on **AfriTeVa V2**, a transformer-based model specifically designed for African languages, and fine-tuned on over 100K curated translation pairs. --- ## ✨ Key Features ### 🔄 Intelligent Translation Pipeline - **Multi-Output Generation**: Produces 4 translation variants using different decoding strategies (beam search, sampling, etc.) - **Quality Scoring**: Automatically ranks translations based on length ratios, vocabulary diversity, and coherence - **Hallucination Detection**: Identifies and flags potentially unreliable translations - **Context-Aware Chunking**: Preserves sentence boundaries when processing long texts (80 tokens per chun …