Machine Learning Approach to English-Afaan Oromo Text-Text Translation: Using Attention based Neural Machine Translation
Ebisa A. Gemechu, G. R. Kanagachidambaresan · 2021 4th International Conference on Computing and Communications Technologies (ICCCT) · 2021
In this paper, we present a Neural Machine Translation (NMT) approach for English-Afaan Oromo text translation. NMT is a machine translation technique that applies an artificial neural network to predict the probability of a sequence of words. It is enhanced by Recurrent Neural Networks (RNN), also called the encoder-decoder networks. There have been some noticeable challenges with English-Afaan Oromo machine translation using the conventional rule-based and Statistical Machine Translation (SMT) systems. Text alignment was the main challenge since there are differences in the structure between the two languages. Afaan Oromo is a morphologically rich language, which makes it difficult to develop rules for the syntactic and semantic elements. Our proposal is to develop a machine translation model for English-Afaan Oromo text, using the attention-based NMT technique. This technique overcomes most of the text alignment and rule tagging limitations. The train/test split technique is used to train and test our model. Adam algorithm is adopted for our training optimizations. We used the BLEU score and human rating Likert scale evaluation methods for evaluation. Experiment shows that our model is significantly important in translation with an average BLEU point of 41.62 on the test sets. This result outperforms the previous baseline systems experimented on English-Afaan Oromo.