Unigram Backoff vs. TnT Evaluating Part of Speech Taggers Introduction to Computational Linguistics

Automated statistical part-of-speech (POS) tagging has been a very active research area for many years and is the foundation of natural language processing systems. This paper introduces and analyzes the performance of two part-of-speech taggers, namely the NLTK unigram backoff tagger and the TnT tagger, a trigram tagger. Experimental results show that the TnT tagger outperforms the NLTK unigram backoff tagger. However, the tagging accuracy of both taggers decreases almost by the same number when the taggers are tested on a corpus which differs from the one they were trained on.