A neural network for supervised learning of natural language grammar

Within the framework of the expert information retrieval system, DIALECT 2, we propose a connectionist method for a linguistic morpho-syntactic parser of the French language. The system is based upon a three layered neural network with a recursive sentence structure. This network is in charge of the acquisition of natural language grammatical competence. The learning stage is supervised and distributed into several levels. The learning algorithm uses a measure grounded on an entropic computation. We describe the overall architecture of the system and show the first results obtained with samples made up with sentences from schoolbooks for children who are taught reading.<>