Training and evaluation of TreeTagger on Amazigh corpus

Abstract

Part-of-speech (POS) tagging has high importance in the domain of natural language processing (NLP). POS tagging determines grammatical category to any token, such as noun, verb, adjective, person, gender, etc. Some of the words are ambiguous in their categories and what tagging does is to clear of ambiguous word according to their context. Many taggers are designed with different approaches to reach high accuracy. In this paper we present a Machine Learning algorithm, which combines decision trees model and HMM model to tag Amazigh unknown words. In case of statistical methods such as TreeTagger, this will have added practical advantages also. This paper presents creation of a POS tagged corpus and evaluation of TreeTagger on Amazigh text. The results of experiments on Amazigh text show that TreeTagger provides overall tagging accuracy of 93.19%, specifically, 94.10% on known words and 70.29% on unknown words.