Effect of Parallel Data Processing Model on Bi-Directional English-Khimtagne Machine Translation Using Deep Learning

Abstract

Neural Machine Translation (NMT) is a key application of Natural Language Processing (NLP) that allows text to be translated automatically from one natural language to another without the need for human interaction. The goal of this work was to create a bidirectional machine translation system between English and Khimtagne, an endangered language in Ethiopia, using deep learning techniques. Khimtagne users are expanding, and an effective translation system can help with information interchange and language preservation. The study used a dataset of 11,768 parallel sentences from the Bible to train and test two deep learning encoder-decoder models: CNN with attention and Transformer. The proposed Transformer model obtained BLEU scores of 5.72 for English to Khimtagne and 5.19 for Khimtagne to English translations. While the results show that the approach is feasible, the research's main disadvantage was the relatively small dataset size, which may have constrained the model performances. Further investigation is needed to increase the dataset and investigate more advanced deep learning methods for improving translation quality. Nonetheless, this research is a significant step in closing the language gap and preserving the Khimtagne language. The findings could contribute to the field of Natural Language Processing, where machine translation is an important application for enhancing human-computer interaction utilizing natural languages.