Bi-directional Amharic – Afaan Oromo Machine Translation Using Statistical Approach

Abstract

Machine translation (MT) is an automatic translation from one natural language to another by a computer, without human involvement. The purpose of this study is to develop a bidirectional Amharic- Afaan Oromo machine translation system using statistical machine translation. In this thesis, to explore the effect of morpheme and word level alignment on bi- Directional Amharic-Afaan Oromo statistical machine translation. In order to conduct the study, the corpus was collected from online source such online documents include Old and new Testament of Holy bible and religious documents for both language and corpus preparation which also involves dividing the corpus for training set, tuning set and test set. A total of 14600 sentences are collected. We use 1460 for testing and 1460 for tuning purpose. For language model we used 11680 parallel sentences sentence for both Amharic and Afaan Oromo language. The experiment was conducted using statistical Machine Translation tool moses, MGIZA++ for word and morpheme alignment toolkit, morfessor were used for morphological segmentation for both Amharic and Afaan Oromo language and IRSTLM language modeling tools. Different experiments were carried out after preparing and designing the corpus and the prototype. Experiments were conduct based on the morpheme and word level alignment and results were recorded. The experiments were taken separately. The result obtained for the unsupervised morpheme segmentation based level alignment using BLEU score has an average of 19.77 % accuracy for the Amharic to Afaan Oromo and 16.14 % for the Afaan Oromo to Amharic. For word based alignment, the result acquired from the BLEU Score was 13.84 % for Amharic to Afaan Oromo and 9.72`% for Afaan Oromo to Amharic. This result shows that morpheme level alignment translation performs better than word-level alignment translation.