Word-formation complexity: a learner corpus-based study