Bi-Directional Machine Translation Between Amharic and Khimtagne Using Deep Learning

Natural Language Processing (NLP) is a specialized field of AI that focuses on the interaction between humans and computers in language usage. One practical application of NLP is machine translation, which allows for automated translation without human intervention. In this study, we aimed to develop a machine translation model for Amharic-Khimtagne language pairs using deep learning approaches. Currently, a vast amount of information in Ethiopia is available in Amharic, including rules and regulations, mass media news, religious books, educational materials, and official documents. With the growing number of language users in Khimtagne, developing machine translation systems between Khimtagne and Amharic is crucial for sharing information. In this study, we experimented with four deep learning encoder-decoder models, including LSTM, LSTM with attention, CNN with attention, and Transformer, using 17,153 parallel sentences for each model in bi-directional translation between Amharic and Khimtagne. Our proposed Transformer model achieved a BLEU score of 4.52 for Amharic to Khimtagne and 4.19 for Khimtagne to Amharic translation. However, the lack of sufficient data is a major weakness of this study, and additional parallel corpora are necessary for further research.