Performance Of Neural Machine Translation Tools In Translating Arabic Non-diacritised Terms: A Case Study Of Algerian Family Code

Abstract

This paper aims to evaluate the output of three machine translation tools that rely on the neural approach and artificial intelligence, it seeks to measure the performance of these tools in rendering Arabic non-diacritised legal words extracted from the Algerian Family code written originally in Arabic. To achieve this, a sample of 10 passages containing Arabic non-diacritised terms was selected, the criteria of selecting the passages was based on containing words having multiple meanings according to varying diacritics put on their letters. The samples were given to three different neural machine translation tools to be translated from Arabic to English, the results were then compared according to the error-rate of each tool. The case study showed varying results for the three machine translation tools, it demonstrated the ability of machine translation in disambiguating the correct meaning of Arabic terms from context even with the absence of diacritical marks