sahib-sem/Amharic-Spelling-Error-Detection-and-Correction

# Amharic Grammarly - Spelling Error Detection System and Suggestion ## Objective Amharic Grammarly detects and suggests corrections for misspelled Amharic words using both contextual and non-contextual methods. ## Data Two datasets: 1. **Dictionary**: Words with frequencies. 2. **Corpus**: Amharic text for next word prediction. ## Preprocessing - `dictionary.txt`: Preprocessed and sorted. - Corpus: Normalization using `etnltk` for character and short-form expansion. ## Algorithms 1. **Non-Contextual Correction**: Edit Distance (Levenshtein distance). - Operations: Insertion, deletion, swapping, or replacing characters. 2. **Contextual Correction**: LSTM for next word prediction. - Architecture: Embedding, LSTM layers, fully connected layers. ## Limitations - LSTM vocabulary (46,000 words) vs Dictionary (450,000 words). ## Combining Suggestions 1. Identify misspelled words. 2. Run edit distance for suggestions. 3. Use LSTM for context-based predictions. 4. Combine suggestions based on probabilities. ## Backend Fastapi backend with endpoint `/spellcheck/suggestion` for spellcheck suggestions. ## Resources 1 **Project Report paper** - Report. 2 **Project Presentation Slide** - Presentation slide.