Multi-Dialect Detection for Albanian Language Using Natural Language Processing

The Albanian language is primarily divided into two main dialect groups, Gheg and Tosk, each with distinct regional variations spoken across specific geographical areas in Albania, Kosovo, and North Macedonia. Several studies have explored Albanian dialect classification using diverse methodologies, including corpus-based approaches, geotagged social media analysis, and quantitative linguistic studies. They have demonstrated the effectiveness of machine learning techniques in dialect identification. In our work, we contribute to this growing body of research by training machine learning models for Albanian text classification using TF-IDF feature extraction and the SGDClassifier. Our approach, which incorporates geographical features for dialect classification, proved effective in achieving a more comprehensive classification of Albanian texts. We observed impressive performance in classifying Albania dialects due to the distinct linguistic features and the availability of a larger dataset. Conversely, the classification of dialects from Kosovo and North Macedonia posed significant challenges, primarily stemming from feature overlap and the limited size of available datasets. These challenges highlight the need for more advanced feature extraction methods and present opportunities to enhance dialect classification models across diverse text types and regional variations.