Researchers present a regression-based approach to Arabic dialect geolocation, predicting speaker origin as continuous latitude-longitude coordinates with a median localization error of 481.2 km. The model uses a hierarchical neural architecture and achieves 64.5% and 45.2% accuracy for country and city predictions, respectively.