Machine Translation Post-Editing (MTPE): What is it, and when should you use it?

With the recent AI boom, it wasn’t a surprise that one of the first applications of this new technology was translation. There was already a precedent: Google Translate launched nearly 20 years ago and has spawned countless successors in those two decades since. And while the model for Google Translate has changed over the years, it never shook up the industry quite like AI has now.

But with the advent of machine translation (that’s what us in the translation biz call automatic translation engines like Google Translate, DeepL, even ChatGPT) came lots of confusion and speculation about its actual value and utility to provide high-quality translation.

 What is machine translation?

Machine translation, or MT, is essentially using a computer model to translate text. From a user perspective, it’s pretty straight forward: you put in text in one language, and it spits the same text out in another. But therein lied a problem – early iterations of these models were mostly rule-based or statistical, but anyone who’s ever tried to learn a language knows that there are always exceptions to the rule.

Think of how many meanings the word “plant” could have: are we talking about a green thing that grows from the ground, the act of putting a green thing that goes into the ground, a factory or manufacturing centre, or a spy? These engines didn’t often get these things right at the beginning, and they were really only useful for getting the gist of what was said – figuring out until what time breakfast was offered at the hotel you were staying at abroad or writing a thank you note to your neighbour.

 More recently, machine translation engines have improved, employing neural machine translation (NMT) and large language models (LLMs). Their accuracy has improved significantly now that it no longer is tied specifically to grammar book language rules or statistics, but it still often lacks the human touch required to make a translation “good” rather than just “acceptable.”

And what is machine translation post-editing?

Machine translation post-editing, or MTPE for short, refers to a human-in-the-loop model of employing machine translation engines to provide a finalised translation. It’s a process that recognises that machine translation, even in its current iteration, is fallible, and cannot be relied on solely without further oversight.

MTPE uses machine translation for as a kind of “first draft,” but then passes that draft on to specialised, native, human translators that add that necessary human touch to a translation. The nuance that can only be gained through living and interacting in a culture that just can’t be replicated by machines.

 So when should you use machine translation post-editing?

Machine translation post-editing is far from a one-size fits all solution. Crucially, MTPE should almost never be employed on creative translations, particularly for marketing and advertising content. This is an area where MTPE really falls short, providing “acceptable” results when you really want high-quality transcreation.

However, there are a number of applications where MTPE shines. Texts best suited for MTPE generally share the same criteria:

• large in volume (word count)

• repetitive

• to and from source-rich languages*(source-rich languages are languages that generally have a lot of training data behind them, generally the world’s major working languages)

If these conditions are met, MTPE has the potential to seriously speed up the translation process and significantly reduce costs overall. This, of course, is only if the output MT is high-quality.

Low-quality MT could derail an entire project, introducing costly mistakes and taking much longer than human translation, so picking the best translation engine for the job is key. It’s worth examining the outputs from a few different sources and working with your linguistic team to pick the right ones.

Machine translation, particularly NMT and LLMs, is yet another tool in the translation industry’s belt: when worked with intentionally, practically, and in the correct environment, it can yield great results for everyone.

---

Written by: Kelsey Frick, WIT Account Director