New Results - Improving the lexical coverage of statistical parsers