Google translate and deepL: breaking taboos in translator training

Abstract

Over the past years, whenever we read about using machine translation (MT) we were advised that it was not suitable for regular professional translation. Students were warned to use Google Translate in their assignments or exams, and would be punished if they did. The reason was that MT usually produced a poor outcome whenever it was not applied for controlled language of pre-edited texts within a specific field of expertise, and with post-edition of the target text. This situation led us to avoid using MT. Related to specialized texts, and after a time of training, meanwhile MT systems managed to produce quite acceptable results and help to translate faster. But in recent times, new systems have been developed called Neural Machine Translation (NMT), and the output quality is reasonably good. As translator trainers we asked ourselves: Why not use post-editing of automatic translated texts as a chance for translation pedagogy and focus on diagnosis and therapy of those errors still made by MT as a means to improve command in both, mother and second language?