Amharic-arabic Neural Machine Translation

摘要

Many automatic translation works have been addressed between major European languagepairs, by taking advantage of large scale parallel corpora, but very few research works areconducted on the Amharic-Arabic language pair due to its parallel data scarcity. Two LongShort-Term Memory (LSTM) and Gated Recurrent Units (GRU) based Neural MachineTranslation (NMT) models are developed using Attention-based Encoder-Decoder architecturewhich is adapted from the open-source OpenNMT system. In order to perform the experiment, asmall parallel Quranic text corpus is constructed by modifying the existing monolingual Arabictext and its equivalent translation of Amharic language text corpora available on Tanzile. LSTMand GRU based NMT models and Google Translation system are compared and found thatLSTM based OpenNMT outperforms GRU based OpenNMT and Google Translation system,with a BLEU score of 12%, 11%, and 6% respectively.