Discriminative Training of a Neural Network Statistical Parser

摘要: Discriminative methods have shown significant improvements over traditional generative in many machine learning applications, but there has been difficulty extending them to natural language parsing. One problem is that much of the work on discriminative conflates changes method with parameterization problem. We show how a parser can be trained while still parameterizing according probability model. present three for training neural network estimate probabilities statistical parser, one generative, discriminative, and where model criteria discriminative. The latter outperforms previous two, achieving state-of-the-art levels performance (90.1% F-measure constituents).