Bengali parts-of-speech tagging using Global Linear Model

Abstract

The paper describes an automatic parts-of-speech tagging for Bengali sentences using Global Linear Model (GLM) which learns to represent the whole sentence through a feature vector called Global feature. Tagger has been trained using averaged perceptron algorithm. Performance of this tagger has been compared to Conditional Random Field (CRF), Support Vector Machine (SVM), Hidden Markov Model (HMM) and Maximum Entropy (ME) based Bengali POS tagger. Experimental results show that GLM based Bengali POS tagger has the accuracy of 93.12 %.