Part of Speech Tagging Using Artificial Neural Network for Tigrigna

Abstract

The interaction between human to computer is a daily occurrences. Due to the rapid change of technologies, for example, HCI (Human computer Interaction) and NLP (Natural language processing) streamline the easiness and effectiveness of computer systems that can better communicate and assist human beings. For instance, Part of speech tagging is one attempt in the effort of understanding human language to identify the role of a word in a given context. There are a significant amount of work have been done to address part of speech disambiguation problems towards sematic languages. For example, Part of speech tagging for Tigrigna has been done using hybrid approach; HMM tagger combined with the rule based tagger. Unlike the aforementioned approaches, a neural network approach is an effective method to achieve a better performance towards part of speech tagger for Tigrigna. Therefore, the main objective of the study is developing an effective part of speech tagger scheme for Tigrigna words to attain a better performance. To achieve the previously specified objective, a neural network based part of speech tagging for Tigrigna scheme is proposed. The tagger is used Back propagation learning algorithm in order to train the neural network. In order to evaluate the performance of the proposed scheme, an extensive experiment is conducted using a corpus comprised of the training set and the prepared testing set. The performance of the Tigrigna MLP tagger in the 75% of the dataset and 25% of the dataset is 90.726% and 85.693% using 10-fold cross validation technique respectively while the overall performance of the MLP tagger on the whole datasets which can be used K fold technique is 93.849% which takes a total time about 2 hour to train. The experiment results demonstrate that MLP tagger achieve an efficient performance using prepared testing set. However, this result would improve better if there is more dataset for training and testing.