Kontrastive Fehleranalyse Deutsch-Spanisch der NMT in verschiedenen Textsorten

Abstract

In times of digital transformation, translators need to adapt to machine translation (MT). Initially, this was based on a statistical model (Statistical Machine Translation, SMT), a paradigm in which translations are produced based on statistical models whose parameters are derived from analysis of bilingual text corpora (Koehn, 2009). In 2017, however, Neural Machine Translation (NMT) based on deep learning was launched (Schwan, 2017). This translator uses Convolutional Neural Networks (CNN), artificial neural data and, in case of error, tries again and repeats the sequence until it gets it right (Geitgey, 2016). In this paper we set out to machine translate Germand and Spanish texts from different branches of technical language and finally analyse the errors committed by DeepL. For this purpose, literary, journalistic, technical and official/legal texts were chosen. On one hand, we want to see whether NMT really works as well as we are told and, on the other hand, to see which branches of the translation industry can benefit most from NMT. Another aim is to generate different strategies that can help to overcome the errors caused by DeepL.