01/04/2020

Maximum cost and time optimisation without compromising translation quality? A pure utopia, you mean? Or is it? The advent of neural machine translation has fundamentally changed the answer: by combining machine (pre-)translation and subsequent human post-editing, the translation process now benefits from all three elements: translations are produced faster and cheaper, but to the same quality as a human translation.

How about a 120-page manual translated into Spanish within 3.5 minutes? Or Russian? Or even both? Mankind dreams of being able to translate at the click of a button . Yet, until a few years ago, machine translation (MT) was either expensive to set up or produced results that were more for amusement than for technical communication.

“About three years ago, there was a breakthrough with neural machine translation (NMT). Although it requires large training corpora, it also provides very fluent-sounding and linguistically sound results. There are generic machines available that are not trained. They are cheap to use and, therefore, there is no real barrier to entry,” explains Jasmin Nesbigall, Head of MTPE (Machine Translation and Post-Editing) at oneword. “That’s why many of our customers want to use the technology for themselves so they can save time and money.”

At oneword, every MTPE collaboration always begins with a personal meeting and a precise needs analysis. Then, once the field of expertise, the language directions and many other factors have been established, the appropriate engine can be selected. Since oneword is in contact with a large number of MT suppliers, the advantages and disadvantages of the various machines can be weighed up against each other. We carry out initial test runs in close consultation with our customers and analyse the results in detail. Cost-effectiveness is always a priority, so that an MTPE project does not become more expensive than a human translation.

Texts that are not too long, not too short, not too complicated, but not colloquial either? Some recommendations from providers and experts read more like looking for a needle in a haystack. The experience gathered by experts at oneword suggests that hardly any text can be categorically excluded. “We have already had a number of surprises where texts seemed unsuitable at first but, in retrospect, saved 40% compared to human translations,” says Nikolina Cabraja, who is part of the MT team at oneword and conducts multi-stage feasibility analyses for projects. If there is any doubt, the post-editor assesses the feasibility of the post-editing project in the last instance.

“Without post-editing, machine translation just doesn’t work,” is Jasmin Nesbigall’s credo. “Every day in our MTPE projects we see how small and subtle the machine’s errors are in some cases, but what a huge impact they would have in everyday life.” Therefore, human post-editing, i.e. a post-editor checking the machine translation, is a permanent step in the process at oneword. It requires qualified post-editors, who are supervised and trained by Translation Partner Management and are the most important resource in the MTPE process. “Many of our post-editors were a little sceptical at first about whether, in the long term, the machine would take away their work. But our enthusiasm for MT, fair payment and continuous feedback have turned many a sceptic into a true MTPE fan,” says Nicole Sixdorf, Translation Partner Manager at oneword.

If you would also like to try out MTPE for your company or find out more about machine translation and post-editing, please contact us or request our MTP starter kit, which contains lots of useful information.