الفهرس | Only 14 pages are availabe for public view |
Abstract Machine translation (MT) has been widely used in recent years to increase the speed of translation and to reduce its cost. However, the quality of the translation produced by such systems is still far from being satisfactory. Therefore, the need to evaluate the quality of MT systems has been growing. Many evaluation metrics have been developed recently, whether human or automatic. This study follows one of the most effective ways of evaluating MT output quality known as human error analysis. This paper borrows with some modifications the error typology designed by Farrus et al. (2010). It classifies errors made by MT systems into five main linguistic levels; orthographic, morphological, lexical, semantic, and syntactic. These main categories are divided into subcategories depending on the language pair under study. This paper investigates the translation quality that Google Translate provides in translating four text genres from English into Arabic using error analysis. It aims at gaining a deeper understanding of the error patterns in Google Translate to help developers improve system performance. The findings of the current study assert that GT output quality varies across the different text genres investigated in this paper. The findings also suggest that some error categories account for more errors than others. The syntactic error category accounts for the highest number of errors, whereas the lexical error category accounts for the least number of errors. Moreover, the findings indicate that most errors can affect the intelligibility and readability of the target text |