Authenticating corpora for language learning: a problem and its resolution