MTPE, or what we as a language service provider can expect
Thanks to continuous improvements in the quality of machine translation (MT), especially after the introduction of ‘neural’ machine translation, we as a supplier of language services are increasingly receiving orders, especially from foreign clients, for the proofreading of unmodified machine translations (in the terminology of project managers ‘MTPE’, or the post-editing of machine translations). Let’s take a look at what’s actually going on.
Modern technology has been successfully used for a number of years now to improve and speed up the entire translation process. So far, this has mainly been accomplished by translation memories and terminology databases that most language service providers create for their clients and use when preparing translations. And because identical or repetitive parts of the text make up a significant portion of translations for our clients, we can provide them a significant discount, thus allowing customers to reduce their costs with our help.
But what about the parts of the text that are new or are missing from the translation memory? Until now, these were treated from the perspective of savings as text that needs to be translated from scratch or edited considerably, i.e. for the standard price. And the machine translations that translators had available were seen as a useful aid or a bonus for them.
However, customers naturally noticed the increased quality of ‘raw machine’ translations and are now requesting with increasing frequency the simple editing of the output of machine translations (post-editing). And since it is assumed that such orders require less work, customers are demanding post-editing for a rate lower than standard translations.
But is the post-editing of a machine translation really less time consuming to qualify as a unique product?
It depends on a number of factors. Experience shows that although the quality of machine translation has improved dramatically, especially recently, the quality in fact continues to fluctuate considerably. It depends, for example, on the language combination (machine translators handle some language pairs very well, while others produce disappointing results), the field of the specific text or even on the style in which the original text is written. And then again, in some cases our jaws literally drop at what machine translators are able to do with a text written with poor grammar, spelling mistakes or a complicated sentence structure lacking logical connections.
So, what’s the best way to approach the increasing demand for the post-editing of machine translations?
We recommend make decisions on a case-by-case basis, ideally with a test sample, as even a small piece of text can be very indicative of overall quality.
What is your experience with MTPE? We’d be delighted to hear from you.
Author: Lukáš Utíkal
