Implementation of the Serbian Language POS Taggers Using the NLTK Library

This paper describes research on different methods for implementing part-of-speech taggers for the Serbian language. Experiments were conducted on the SrWaC corpus, which consists of 10 million tokens collected from. rs web domain. Algorithms from NLTK library were used for creating tagging models. The tagset was modified and adapted for semi-deep tagging. The experiments involved preprocessing of the corpus to correct tagging mistakes and reversing the word order in sentences to obtain a reversed corpus, which is used for creating backwards tagging model. To achieve greater accuracy, a tagger that combines n-gram taggers was implemented using the NLTK library. However, the best result achieved during tagging was with the Perceptron tagger, which reached 95% accuracy. The goal of this work is, among other things, to ensure that tagging is fast, as the intention is to use the best models for a syntactic analysis tool for the Serbian language.