Abstract:
Being under the constant threat of “publish (in English!) or perish”, academics use public machine translation systems in order to translate articles intended for publication in international journals. Though free and accessible, MT systems do not always produce high quality output. Target texts can contain errors ranging from minor to very severe. This interdisciplinary paper focuses on the results of a small-scale study. A political science academic text was translated from Romanian into English using three neural MT systems: Bing Microsoft Translator, DeepL and Google Translate. Errors were manually annotated in CATMA, a digital tool, taking into account the fluency, accuracy and fitness for purpose of the target texts. Translators, MT systems developers and academics alike can profit from these findings by understanding what kind of errors to expect – and eventually –, correct and eliminate when automatically translating academic texts from Romanian into English.