Anybody know of work to minimize a test corpus for grammar-based parsers?