Measuring Harmful Representations in Scandinavian Language Models