Neural machine translation (NMT) is a revolutionary approach leading to substantial improvements in translation quality. It is based on machine learning and does not require a degree in linguistics to be understood, nor extensive programming skills to experiment with it. This presentation and its accompanying website provide a step-by-step introduction on how a high quality NMT system based on Google’s transformer architecture can be constructed.
Participants will learn from a practical perspective how neural machine translation (NMT) systems work and how they can build their own system on a PC using the freely available and well established Marian NMT toolkit.
Extensive experience in using computers will be assumed. However, prior knowledge of machine learning, machine translation and programming skills are not required.
Reinhard Rapp holds faculty positions at the University of Mainz and at Magdeburg Stendal University of Applied Sciences and is currentlly on leave for conducting a research project on semantics-based machine translation at the Athena Research and Innovation Center in Athens, Greece.