Genetic optimization of NN topologies for the task of natural language processing

Several researchers have successfully used artificial neural networks (NN) to process natural languages. The way in which the neurons of these networks were connected to each other was based on notions of how it was that the NN should attack the problem at hand. How to connect neurons in order to guarantee optimal performance for any one task is still an open question. At the same time, researchers have found that this question can affect performance. This paper presents details of how I have configured a genetic algorithm (GA) to search for good NN topologies for a particular natural language task. This GA makes no assumptions of how the task should it be attacked. Results obtained by the system are also presented.