The Automatic translation systems (ATS) for translation text have extent widely in recent years. The ARS developed to correct several types of text errors explained by the Mossop's prototype such as spelling, typographical, syntactic, semantic, word, and formal ones. The ARS need a large amount of data training in its forms. There is a shortage in German-Arabic datasets for translation and revision purposes. Building dataset is the most time-consuming and the most important part of the text translation process. We make an effort to analyze and work on this large amount of data Sentences, and the form of text free dataset on the ARS, most efforts focus on German and Arabic data. Despite the increase in the number of Arabic, users and the increase in Arabic content on ARS. Therefore, in this paper, Arabic dataset built to use in text translation purpose. This research offers the German-Arabic dataset from the Taxonomy of errors in postediting text for growth the ARS. Our dataset gathered from A Game of Throne saga in German (GR) and Arabic (AR) saga. Our dataset consists of 65,000 bilingual sentences collected from Text. The most significant penalties of this research were the Mossop's prototype terminates to explain all errors; and the prototype had to be lengthy in demand to include the Consistency. Finally, human evaluators were employed to grade the quality of ATS outputs and to revision them. We used a Rapid Miner tool to evaluate the performance of our dataset, the dataset accuracy of 95.12%.