English > Welsh
Machine Translation
Using Marian NMT

The Language Technologies Unit has developed machine translation tools that use the Marian NMT framework. The framework uses modern neural technology to translate texts. Click on the link or image to try it out:


We have provided a special GitHub project for Marian NMT here:

techiaith/ docker-marian-nmt

The project contains code, scripts and documentation which enables you to host your own Marian NMT local machine translation system.

To run a local machine translation (MT) service successfully, you will need a computer with a supported NVIDIA graphics card as well as software packages such as Docker and Python.

The MT system uses translation models that have been specially trained by the Language Technologies Unit for translating from English to Welsh.

There are two methods of using machine translation with this project. The first, through a local API service, that provides the ability to integrate with translation tools (if the software allows). The second, through a live demo website that is also publicly available at:


At the moment, it is possible to translate the text of two specific fields, namely the field of Health and care and the field of Legislation.

Welsh <> English

Machine Translation

Using Moses SMT



Moses-SMT is an open source translation engine developed mainly at Edinburgh University. Research and development for Moses-SMT was funded by the European Union.

It’s a powerful product which allows you to develop your own machine translation systems by training them with corpora of paralell texts of translations.  It allows you to create engines that are specifically adapted to the needs of your translation projects, ensuring that they produce high quality translations.

Moses-SMT is useful in situations where translation memories are already being used successfully to save time and money. For many translators, Moses-SMT offers not only improved translations compared to ordinary translation engines, but has also proven that it’s possible to be far more productive using a post-editing approach than by using only translation memory systems. For this reason, the use of Moses-SMT has recently increased substantially within the international translation industry.

Using the Language Technologies Portal it’s now possible for anyone, but particularly those working in the Welsh <> English translation industry, to benefit from using Moses-SMT in their projects, product or translation services.

We have developed these resources in such a way that you should not require much technical knowledge or ability to use them. To see how you can get started, go to our project page on GitHub for Welsh<>English Moses-SMT :