gitdeeper13/moroccan_nlp

Linguistic Resources and Models for Moroccan Darija and Arabic # moroccan_nlp ### Natural Language Processing: Linguistic Resources and Models for Moroccan Darija and Arabic **DarijaBERT · Baseline Classifier · Linguistic Corpora · AI for Under-Resourced Languages** --- ## 📌 Overview **moroccan_nlp** is a comprehensive project dedicated to developing linguistic resources and Natural Language Processing (NLP) models for Moroccan Darija and Arabic. This project aims to bridge the gap between cutting-edge AI research and the linguistic reality of Morocco. > *"Building Moroccan AI, one word at a time."* ## 🗂️ Table of Contents - Overview - Key Features - Core Model: DarijaBERT - Datasets - Model Performance - Project Structure - Quick Start - Installation - Usage Examples - Platforms & Mirrors - Clone & Download - Citation - License - Author ## ✨ Key Features - **DarijaBERT Integration**: First BERT model for Moroccan Darija (0.2B parameters, ~100M tokens) - **Baseline Classifier**: Keyword-based classification with 100% accuracy on test data - **Linguistic Resources**: Curated datasets for Darija and Arabic - **Open Source**: MIT licensed, available on PyPI - **Reproducible Research**: Full infrastructure with Zenodo, OSF, and Internet Archive ## 🧠 Core Model: DarijaBERT **DarijaBERT** is the first open-source BERT model for the Moroccan Arabic dialect, developed by AIOX Lab and SI2M Lab (INSEA). | Property | Value | |----------|-------| | Architecture | BERT-base (without NSP) | | Model Size | 0.2B parameters | | Training Data | ~3M sequences, 691MB, ~100M tokens | | Sources | Stories, YouTube comments, Tweets | | Vocabulary Size | 80,000 | | Monthly Downloads | 1,296 | | License | Research use only (contact: dbert@aiox-labs.com) | ### Loading the Model ```python from transformers import AutoTokenizer, AutoModel tokenizer = AutoTokenizer.from_pretrained("SI2M-Lab/DarijaBERT") model = AutoModel.from_pretrained("SI2M-Lab/DarijaBERT") ``` ### Fill-Mask Example ```python from transformers import pipeline …