Updated: Mar 25, 2021
Using Machine Translation, of course.
You may or may not be aware of it, but that is what many translation agencies are doing already.
This helps them and their clients:
· reduce cost
· speed up workflows
· get better translation quality (surprisingly!)
MT works especially well in some categories of translation services, such as general business, administrative, corporate and legal. Anything written in plain language gets translated with amazing accuracy, too.
Machine Translation Post-Editing (MTPE) is now standard process
For sure, MT has been getting a bad rap for years, especially among linguists.
Many did not want to have anything to do with it until recently.
I remember receiving regular MTPE requests over the past 15 years or so, and I would decline them every time. Because having to rewrite everything would have taken me much longer than doing the translation from scratch.
But things have changed and the tipping point was reached sometime around 2019. Adaptive Neural Machine Translation has been improving at an incredible pace.
MTPE has now become a standard process in most agencies, whether as a way of producing lower-cost “lightly” edited machine translations of low-key texts, or as part of a speedier translation process involving full human revision.
Based on our own calculations, MT may speed up high-quality translation delivery by up to 50 percent.
How is this possible?
Engines learning from online corpora and human inputs
MT engines are now able to beat some human translators even with terminology and wording.
They also learn from millions of human inputs every day as people correct MT outputs on public domain texts.
They readily produce what might take a human translator hours to finesse. Interestingly, the engine will often stumble over some seemingly simple words, making silly blunders.
These, however, are easy to correct.
Linguists then act as copy editors (or in some cases as copywriters) to make it perfect.
(Note: You must be careful not to choose the ‘human-in-the-loop’ option when applying MT to sensitive documents; otherwise, your texts may end up in the engine’s memory to be accessed by others).
Accurate and up to 50 pct faster
Here’s our first 100-word test from French into English.
The output is not perfect, but it is perfectly editable.
The revising linguist would normally edit it in the system to get it right in the Translation Memory and then generate the translation.
We did it the other way round here, so you could see the Track Changes markup.
Here it is.
It is obvious that the process may be at least 50 percent faster than it would be sans MT.
And here the output after the edits.
Again, similar eye-pleasing results.
Conclusion: Adaptive Neural Machine Translation has become a constantly evolving efficient tool for multilingual scaling, completely revolutionizing the entire language services industry and linguistic roles.