A trigram language model to predict part of speech tags using neural network

Abstract

This paper presents a novel approach of part of speech tagging using neural networks for Punjabi language. To the best of our knowledge neural networks have never been used for the prediction of part of speech tags for Punjabi language. In this paper, a multi layer perceptron neural network tagger with fixed context length has been proposed for tagging of punjabi text. The learning algorithm used for the proposed tagger is error back propagation learning algorithm. The tagged corpus was divided into training and testing data with a randomize function. A feature vector was generated from training data by considering neighboring context for the current word. Trigram model has been used for generating this feature vector for every word in the training data. An overall accuracy of 88.95% is achieved from the tagger. Results shows that the proposed neural network based tagger performs better than existing taggers for punjabi. © 2013 Springer-Verlag.