Certification and Cleaning up of a Text Corpus: Towards an Evaluation of the "Grammatical" Quality of a Corpus