Improving the identification of verbs

Abstract

In automatically assigning part-of-speech tags to scientific text, we find a high error rate when tagging main verbs. To reduce the occurrence of this serious error, we have created a neural network to search for main verbs that have been mis-tagged by a rule-based tagger. In this paper we describe our efforts to evolve the connection weights for the neural network, and to fractally configure another neural network for the same task.