Arralle21/Somali-Homographs-NLP

# Somali Homographs NLP This repository contains the computational analysis code for the Springer-published paper "Creating and Analyzing a Dictionary-Based Lexical Resource for Somali Homograph Disambiguation" by Abdullahi Mohamed Jibril and Abdisalam Mahamed Badel (2025). The project features the first-ever Somali homograph dataset with 1,592 unique homographs extracted from the Qaamuuska Af-Soomaaliga dictionary. Features The analysis includes comprehensive statistical distributions across the 26-letter Somali alphabet, semantic similarity measurements using TF-IDF and sentence transformers, and machine learning clustering with evaluation metrics. The codebase generates high-resolution visualizations for publication and provides tools for homograph frequency analysis, meaning distribution studies, and semantic clustering evaluation. ``Repository Structure`` Dataset/ - Contains the Somali homographs CSV file with definitions Plots/ - High-resolution visualization outputs (600 DPI) for analysis results somali__homographs.py - Main analysis script with all computational functions Somali__Homographs.ipynb - Jupyter notebook version for interactive analysis Key Analysis Components ``Statistical Analysis`` Distribution of homographs across the 26-letter Somali alphabet Frequency analysis of meanings per homograph (average 2.5 meanings per word) Word length distribution and most common definition terms Identification of most ambiguous homographs by starting letter ``Semantic Analysis`` TF-IDF vectorization for semantic similarity measurement Sentence transformer embeddings using paraphrase-MiniLM-L6-v2 model Cosine similarity calculations between homograph definitions Clustering evaluation with Silhouette, Calinski-Harabasz, and Davies-Bouldin indices ``Machine Learning`` K-means clustering of homograph definitions t-SNE visualization of semantic clusters Comparative analysis of different embedding methods Performance evaluation across multiple clustering algor …