fuadmuhe12/Amharic-spell-checker

# Amharic Spell Checker (Interpolated N-gram Model) A robust, context-aware spell checker for the Amharic language. This project uses a **Linear Interpolated N-gram Model** (Trigram + Bigram + Unigram) to suggest corrections that fit the context of the sentence. ## 🚀 Key Features * **Interpolated Smoothing**: Combines Trigram, Bigram, and Unigram probabilities to handle unseen contexts gracefully. * **Efficient Edit Distance**: Uses optimized Damerau-Levenshtein distance to find candidate words. * **Amharic Preprocessing**: specialized normalization and tokenization for the Ethiopic script. * **Modern Interface**: Includes a **Streamlit Web App** and a **Typer CLI**. * **Type-Safe**: Fully type-hinted codebase. ## 📦 Installation This project is managed with `pyproject.toml`. ```bash # Clone the repository git clone github.com cd amharic-spell-checker # Install dependencies pip install -e .[dev] ``` ## 🛠️ Usage ### 1. Training the Model Before using the spell checker, you must train the N-gram model on a corpus. ```bash python scripts/train.py data/amharic_corpus.txt --output-path models/ngram_model.pkl ``` ### 2. Web Interface (Streamlit) Launch the interactive web demo: ```bash streamlit run src/amharic_spell/app/web.py ``` ### 3. Command Line Interface (CLI) Check a single sentence: ```bash python src/amharic_spell/app/cli.py "የቤት ውስጥ ስራ የሴቶች ስራ ብቻ ሳይቻሆን" ``` ## 🏗️ Architecture ```mermaid graph TD User[User Input] --> Tokenizer[AmharicTokenizer] Tokenizer --> Detector{In Dictionary?} Detector -- Yes --> Correct[Correct Word] Detector -- No --> Generator[Candidate Generator] subgraph "Correction Engine" Generator -->|Edit Distance| Candidates[List of Candidates] Candidates --> Scorer[N-gram Scorer] Model[(Interpolated LM)] --> Scorer Scorer --> Ranker[Rank Suggestions] end Ranker --> Result[Final Suggestions] ``` ## 🧪 Testing Run duplicate tests to ensure reliability: ```bash pytest `` …