Fast Syntactic Analysis for Statistical Language Modeling via Substructure Sharing and Uptraining