Amharic-Kistangigna Bi-directional Machine Translation using Deep Learning

The main goal of this work is to develop bi-directional machine translation between Amharic and Kistanigna. Due to an increase in the number of language users, to address the issues of the endangered of the Kistanigna language and to increases the content of the language in web, it is essential to develop machine translation between Kistanigna and Amharic language. In this study the translation is implemented using neural machine translation. We conduct the experiments using LSTM, Bi-LSTM, LSTM + attention, CNN + attention and Transformer encoder-decoder model. We considered training time, memory usage, and BLEU score when proposing an optimal model. Finally, we suggested the morpheme-based bidirectional machine translation using Transformer with BLEU scores of 21.31 and 22.40 for Amharic-Kistanigna and Kistanigna-Amharic translation respectively. The major weakness of the study is unavailability of enough dataset to conduct an extensive experiment. As a result, there is a need to prepare parallel corpora for conducting similar research.