Die Transformer-Architektur für Systeme zur neuronalen maschinellen Übersetzung – eine popularisierende Darstellung

This paper provides a gentle introduction to the Transformer architecture for neural machine translation (NMT) systems for audiences from the fields of translation and/or LSP communication (studies). Introduced in 2017, the Transformer has become the standard architecture for NMT systems, replacing the previous architecture based on recurrent neural networks (RNN). Since NMT has a growing impact on the practice and study of translation, there is a certain need for stakeholders in these fields to acquaint themselves with the inner workings of this technology. The present paper is intended to serve as a point of access to the technical foundations of NMT.