Morpheme Based Bi-Directional Machine Translation The Case of Ge’ez to Tigrigna

Abstract

Both Ge‟ez and Tigrigna languages, which the native Ethiopian languages, are morphological rich and complex for bi-directional machine translation. To overcome this machine translation problem, this study explored the effect of morpheme-based translation unit for bidirectional Ge‟ez and Tigrigna languages. The corpus was taken from Ten Bible Books that contained 384 that contained 9189 verses. The corpus was used both for developing pre-trained model and for validation. Accordingly, to train the morfessor, 12173 simple Ge‟ez and 16708 Tigrigna words were taken from SQlite database. Explicitly, from the total of 7290 verses data, 80%, that is 7290 Verses were used to develop the pre-trained model and 20% which is 1899 Verses were used for testing or validation purposes. we used Mosses for translation process, MGIZA++ for alignment of word and morpheme, morfessor and IRSTLM techniques for the language modeling. After preparing and designing the prototype and the corpus, different experiments were conducted. BLUE score which is standard for automatic machine translation evaluation was used to measure how much of the system output is correct. Experimental results showed a better performance of 9.23% and 8.67% BLEU scores using morpheme-based from Geez to Tigrigna and from Tigrigna to Geez translation, respectively. That is, it was found out that the model or the system output was correct. Regarding the BLUE metrics evaluation tool, it was also found to show proper validation scores or results. As to the alignment challenges, many-to-many alignment is the major challenge. Hence, there is a need to conduct further research to handle the issue of many-to-many alignment challenge.