Adapting Large Language Models for Underrepresented Languages