Learning incremental syntactic structures with recursive neural networks

摘要: We develop novel algorithmic ideas for building a natural language parser grounded upon the hypothesis of incrementality, which is widely supported by experimental data as model human parsing. Our proposal relies on machine learning technique predicting correctness partial syntactic structures that are built during parsing process. A recursive neural network architecture employed computing predictions after training phase examples drawn from corpus parsed sentences, Penn Treebank. results indicate viability approach and lay out premises generation algorithms processing more closely These may prove very useful in development efficient parsers have an immediate application construction semiautomatic annotation tools.