Morphological Segmentation Using Neural Networks For Afaan Oromo

Abstract

Natural Language Processing (NLP) concerns with computational processing of natural languages in order to provide a products as computers interact linguistically with people in ways that suit people rather than computers. Morphological segmentation is one of the applications of natural language processing that studies the use of computer programs and software to segment words to their morphemes. Morphological segmentation is used as components in many applications, specially machine translation, spell-checker, Part of Speech Tagging (POS) tagging. Several researchers have applied machine learning approaches for Afaan Oromo morphological segmentation while no research have used artificial neural networks for morphological segmentation task. Artificial neural network is subset of machine learning which inspired by the structure, processing method and learning ability of a biological brain. The processing of multiple data inputs is done by different machine learning algorithms. Hence, Neural Networks have the ability to learn by themselves and produce the output that is not limited to the input provided to them. Morphological segmentation using neural networks have been developed for languages such as English. Thus, the main aim of study is to development of a morphological segmentation using neural networks for Afaan Oromo. In order to achieve the objective of this research work, a corpus is collected from different sources such as Books, Newspapers of Afaan Oromo and prepared in a format suitable for use in the development process. We have used corpus of size 50,200, which we have been developed. From this corpus we have used corpus of size 40,160 for training and 10,040 for testing of our work. From the experiments F-score achieved was 97.48%, 98.33%, 98% using Bidirectional Long Short Term Memory, Long Short Term Memory, and Recurrent Neural Networks respectively. In conclusion, the accuracy of the Afaan Oromo morphological segmentation using neural networks were promising than baseline experiments. To improve the performance of the model increase number of training data were recommended for future works