18/03/2022

Software localisation and MTPE

In the right conditions, machine translation can make a significant contribution to localising effective and user-friendly software products in any desired language. We explain how this works best.

The worldwide sale of software products largely depends on the fact that they can be easily understood and used in every language and every culture. Coherent localisation ensures this if the parameters are correct during preparation and implementation.

The use of machine translation (MT) promises greater efficiency and faster multilingual results. The software industry can actually benefit greatly from this. This is because, especially in the case of software documentation, online help, manuals and support texts or even resource files and strings, machine translation achieves a level of quality that MT systems in other sectors and specialist areas usually cannot match.
The Intento Report on machine translation already proved this in 2020: “The highest MT quality is available for computer software, legal services and telecommunications, with software strings and documentation, support content, policies, processes and procedures being the most accessible content types.”

With standard short product cycles, agile sprints and extensive documentation, it is a distinct advantage to have appropriate translations available quickly and cost-effectively. Another advantage is also that industry- or domain-specific MT systems can be used for software translation, which are available practically ad hoc and without MT training. In many cases, even the very low-cost generic “all-rounder” MT systems can be used, which already cover a variety of language directions and subject areas in good quality and make initial investments in machine systems and training time equally obsolete.

However, in a professional context, a closer look reveals limitations and risks (article only available in German) of machine translation of software products. First and foremost, there is potential for errors in localisation and usability.

For example, an MT system cannot take into account any context in the form of culturally typical, legal or other user conventions, but considers each sentence on its own “context-neutral” from point to point.
Unlike human translation, where the software product or reference material is usually available and it is possible to research exactly what is meant by a particular word and its relationship, an MT system has to “guess” at this point. Without coherent context, it cannot create a consistent translation. However, this is indispensable for menus, commands, functions etc. on the user interface, a universally comprehensible choice of words and convenient usability.

As a general rule, machine translation is only suitable to a limited extent for user interfaces, panels, displays and human-machine interfaces (HMI), which are both very complex, very low-context and often space-limited.

The results of machine translation are respectable and getting better and better. Depending on the software content, user interfaces are therefore also increasingly pre-translated with the help of MT systems. However, without an overview and contextualisation, potential errors and sources of error also remain considerably high. (There are so many that we even have our own guide (only available in German) on how to avoid them.)

This is especially true for MT-affine software localisation: The user interface and the manual must match in terms of content and terminology. An MT system cannot do this, because it is continuously calculating new probabilities. Regardless of whether it has had sector-specific training or generic, it usually remains inconsistent.
If an MT system were to be specially trained for this purpose, i.e. if a customised MT system were developed, it would need very large training corpora. However, the time and cost involved in creating and using them is up for discussion for most software providers.

The big difference and a step towards consistent and conclusive results is marked by the previously hidden, small syllable “before”. (Had you noticed?) In professional use, MT is used for pre-translation and is completed by the addition PE to Machine Translation + Post-Editing, MTPE for short.
The PE refers the decisive phase in which machine pre-translations are corrected, supplemented and standardised by trained translators and linguists – the post-editors – in accordance with specific specifications and DIN ISO 18587.

Software localisation + MTPE strategy (source: oneword GmbH)

Machine (pre-)translation often offers enormous cost and time advantages, especially for software and software products as well as for the entire product marketing. When translating user interfaces, however, effort and (economic) benefit must be weighed up carefully: in some cases, it is advisable to have a human-translated user interface – as a kind of starting or reference object for translation memory and terminology compilation – which is then consequently used for further MTPE applications, such as manuals and documentation. This produces a correct, accurate and cost-effective result that is comparable to a human translation.

Every software product and every localisation has individual prerequisites and requirements. To capture these, a needs and feasibility analysis is first required – and based on this, the right strategy for MTPE-supported software localisation with optimised processes, costs and lead times.

Talk to us. Our experts are available to develop your future localisation strategy together.