... model outperforms the n-gram model for bothtypes of classification, and achieves the same per-formance as the discriminative system of Foster etal. (2008), which is state -of- the-art for this data ... University.968Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics, pages 959–968,Jeju, Republic of Korea, 8-14 July 2012.c2012 Association for Computational LinguisticsLarge -Scale ... conditioned on contexts consisting of previously generated treelets (rules, nodes, etc.).Formally, let T be a constituency tree consisting of context-free rules of the form r = P → C1· · · Cd,where...