How to post-edit machine translations for fast, reliable results

Published on August 12th, 2021

Machine translation instantly converts content into foreign languages, but the technology isn’t reliable enough for commercial purposes. Last year, Facebook’s machine translation system turned Chinese President Xi Jinping’s name into an obscenity for the world to see – an incident the social media giant blamed on a “technical issue”.

Of course, Facebook’s translation technology isn’t designed for commercial use, but this unfortunate example is as bad as translation gaffes get and also, a perfect one because it shows why post-editing machine translations is so important.

What is machine translation post-editing?

Machine translation is great for speed but not so reliable for quality and accuracy. In the case discussed above, Facebook explained that the Chinese president’s name hadn’t been inputted into the database that translates Burmese into English.

This touches on the big vulnerability of machine translation – it is only as good as the data it has access to and the algorithms’ capability of translating accurately.

The quality of these translations varies greatly, depending on the languages in question and the linguistic complexity of the content involved. The post-editing of machine translations maintains the quality of automated translation to bring it up to the level of professional human translation.

Essentially, post-editing picks up any errors and fixes them before the finalised translation is delivered.

The aim is to take advantage of the speed benefits machine translation has to offer while correcting its tendency for inaccuracies. So, if algorithms inadvertently mistranslate the name of a world leader into profanity, machine translation post-editing can save companies from such unwanted embarrassments.

Fast translation without the compromises

Post-editing is important for any instance where you’re using machine translation for commercial or public purposes. Generally speaking, you’ll use this technology in instances where a rapid translation is needed and human linguists cannot manually produce the content quickly enough.

In some cases, this could be large pieces of content with short turnaround times. In others, it may be demanding scenarios where instant translation is needed for live commentary, live broadcasts or other real-time events.

In these situations, post-editing is vital for ensuring accuracy and avoiding any damaging translation errors.

Broadly speaking, there are two types of post-editing:

  1. Light post-editing (LPE): Where editors only modify translations where absolutely necessary – to maximise speed or cost-effectiveness while avoiding the worst issues.
  2. Full post-editing (FPE): Where content is thoroughly reviewed and edited to maximise quality and consistency.

LPE is more suitable for those live instances or high-demand scenarios where speed is the priority, while FPE is the preferable option when you want the content to feel as if it was originally created in the target language.

It is important though to consider that, with FPE, there comes a point where you have to ask whether machine translation should be used as the primary solution if quality is the priority, as opposed to using professional translators, to begin with.

It all comes down to striking the right balance between time, cost and output quality – aspects that should be assessed on a project-by-project basis, and which we advise our clients on.

If you’re currently working on projects that you wish to machine translate, feel free to check out our MT plus and AI plus solutions – or get in touch with one of our MT consultants today.

Posted on: August 12th, 2021