Effect of Diacritics on Machine Translation Performance: A Case Study of Yemeni Literature

Abstract

Many Arabic texts are written without diacritics. However, in some contexts this raises the high level of homography and in turn presents difficulties for
machine translation programs. Homographs are words which are spelled identically but have different meanings and are mostly pronounced differently. To avoid the problem of homography, words require to be diacriticized. Thus, the main objective of the study is to assess the ability of machine translation (henceforth MT) in rendering diacritical words from Arabic into English with special reference to translating Yemeni literature into English. This study will also compare the translations of three MT programs, namely, (Reverso, Systran Translate and Free Translation Online) to find out which program is close to the original meaning of the source language texts. Further, the study aims to identify some causes that stand behind errors of translating diacriticized words that result from the mentioned programs. To achieve these aims, descriptive, analytical and comparative methods were followed by the researcher. Thus, the three common and modern MT programs, Reverso, Systran and Free Translation Online were selected to translate some diacriticized words. Then, some excerpts with their contexts were taken from the two Yemeni works, The Hostage (Ar-rahinah) (???????) by the Yemeni famous writer Zayd Muttee Dammaj and the Yemeni book Yemeni Wealth from Popular Proverbs ?????? ??????? ?? ??????? ??????? by the Yemeni writer Muhammad Al-Adimi. The chosen samples were inserted into these MT programs to be electronically translated and then analyzed and discussed qualitatively and quantitatively. The study has concluded that MT encountered the problem of diacritics in Arabic texts; as a result most of the time MT programs failed in recognizing diacritics on letters. Thus, most of the programs' translation results were incorrect and did not concord with the original meaning. It can also be concluded that Free Translation Online program produced the least errors of the three programs and Systran mistranslated all the diacriticized excerpts. These errors can be attributed to absence of programs which contain the diacritic system of Arabic.