Describir: Measuring the semantic headedness of English blends with token-based semantic vector space modeling: a corpus-based study.