ATTENTION BASED ENGLISH-AFAAN OROMO NEURAL MACHINE TRANSLATION MASTER'S THESIS

 Advisor Mr. Jabesa Daba (Assist. Professor)  Neural Machine Translation (NMT) is a method for learning automatic translation using a single large Neural Network. It represents the entire translation process from beginning to end without the need for extra components like Statistical Machine Translation systems. The translation performance of Neural Machine Translation has been encouraging. It is now the accepted method in machine translation research. In this Research work, we applied Neural Machine Translation for English-Afaan Oromo Language pair. In particular, Attention-Based Mechanism was used for developing the machine translation system. Attention Mechanisms are being increasingly used to improve the performance of Neural Machine Translation (NMT) by selectively focusing on sub-parts of the sentence during translation. This capability also makes the Attentional Model superior in translating longer sentences. In order to achieve our objective, we implemented two systems the first system uses Recurrent Neural Network (RNN) based model with use of Long Short-Term Memory (LSTM) type of Recurrent Neural Network architecture using encoder-decoder based language modeling and second, we implemented our Attention Based Recurrent Neural Network (ARNN) model by using Bi Directional Recurrent Neural Network model for English-Afaan Oromo language pair. Also, we collected parallel corpus for English-Afaan Oromo Language pair from different sources and divided into training and testing sets. We have used 80% of total dataset for training and 20% of total dataset for testing. After training and testing these systems on corresponding training and testing datasets, the Attention Based Recurrent Neural Network achieved 62.14 BLEU score on translation from English to Afaan Oromo and Recurrent Neural Network based approach achieved 40.48 BLEU score on translation from English to Afaan Oromo translation. As the results of the experiment confirmed that Attention Based Recurrent Neural Network model works better than non-Attention Recurrent Neural Network model system with increase of sentence length. Therefore, when longer sentence is the candidate, Attention Based Recurrent Neural Network model is the better choice than Recurrent Neural Network model. Keywords: Machine translation, English-Afaan Oromo machine translation, recurrent neural network (RNN), Attention Based Recurrent Neural (ARNN), bi-directional RNN machine translation, long short term memory (LSTM)