Translating with human and machine power

Artificial Intelligence has allowed us to boost productivity in machine translation. So where to from here? Which way to go with regard to pre-editing and training with one’s own data? What we need is the expertise of technical communicators.

Text by Rachel Herwartz

Inhaltsübersicht

Image: © baona/istockphoto.com

In Neural Machine Translation (NMT), a number vector is generated for each word in the input sentence (“input word”). During the machine translation process, an encoder enriches each input word with contextual information from the entire input sentence. A decoder generates the output words and builds the output sentence word by word. The probability of each following word results from the words of the input sentence and all previously generated output words (Figure 1). 

Figure 1: Encoder decoder architecture

The success of this translation method is based on the use of artificial neural networks. In the brain, nerve cells (neurons) are connected to each other by synapses, thus creating “neural networks”. 

Artificial neural networks (ANNs) are programs that learn what the expected outputs are for certain input values by linking data during training. This is how AI systems can ...