Reducing Confusion in Active Learning for Part-Of-Speech Tagging