A neural network that learns to do hyphenation

Abstract

Summary form only given, as follows. Hyphenation of German words is a highly irregular problem. Existing solutions for automatic hyphenation are not very satisfying. A 'sequential network' was applied to this problem. The training algorithm was standard backpropagation. The network was trained with a collection of 1000 German words together with their correct hyphenation. In subsequent tests with unknown words, a correctness of 96.8 percent was achieved. Analysis of the simulation results indicates that with further increases of the training data improvements are still possible. >