Always had the question in my mind whether language specific models would outperform general models