Neural part-of-speech tagging

Abstract

In this work, we propose the implementation of a part-of-speech tagging system using recurrent neural networks. For that purpose, initially we study the theoretical fundamentals of that kind of neural networks. Next, we propose three different architectures in order to disambiguate ambiguous words. Finally, we achieve a system able to disambiguate with a 93.5% total accuracy and with 83.2% accuracy on ambiguous words with the section of Wall Street Journal that belongs to the Penn Treebank corpus.