The Present Use of Statistics in the Evaluation of NLP Parsers