Neural networks for parsing natural language sentences

A neural network approach for parsing natural language sentences in English is presented. It is limited to technical sublanguages, and is based on a multilayer perceptron. The network is able to assign to each word of a sentence its syntactic category, with a low rate of errors. The network is part of a system including pre- and post-processors, and can be run also in parallel to traditional grammar parsing, with both approaches taking advantage of each other to decrease the overall error rate.