Localization lingo clarified: Post-edit machine translation
In this blog, we’ll be diving into the ins and outs of the term “post-edit machine translation” (PEMT).
So, what is it?
PEMT is a translation solution in which source content is first processed through a machine translation (MT) engine and then automatically rendered into another (target) language without the need for human intervention. The direct output from such a system is raw MT. After that, everything these engines crank out is edited and fine-tuned by expert linguists to ensure your original message is rendered correctly in every language.
So machines don’t translate correctly? Actually, sometimes they do. The good ones can even return technically correct content on a regular basis. But sometimes, machines do make mistakes.
Ah – so they produce “non-idiomatic” text (i.e., content that sounds machine translated)? This can be part of the problem, indeed, but machines are becoming more fluent. The real issue is machines sometimes get the meaning wrong, which really hurts.
So if you want to be sure your text does not cause bewilderment, mockery, accidents or diplomatic crises, you should probably not rely on MT alone.
When should post-edit MT (PEMT) be used?
While there isn’t a black and white answer to this question, (check out our webinar on May 30 during which we will drill down on this topic!) there are two main situations in which we recommend using PEMT:
- Quick-turn translation request and moderate quality expectation
Millions of words lined up in need of translation but no time to wait for translators to wade through it all? PEMT can be a great time-saver in a pinch.
Whereas a human translator completes on average 2,000 to 3,000 words per day, a machine can pump out this volume in minutes, and translators can then focus on reviewing it and fixing the mistakes. The better the MT system and the fewer requirements you have regarding your target text, the faster the process can go. This can double the number of words done per day.
- Large-volume translation request and not enough budget for standard translation
Perhaps you’re up to your ears in manuals that need translation, but the amount of words – and work it takes to handle them – is downright daunting. You need to get it done but there’s simply no budget to get high-end translation. If you are prepared to talk about “good enough” quality – consider PEMT for the job.
In one year, we process over 15.7 million words through MT engines for one customer alone. Because they benefit from over 80 percent translation memory leverage, in addition to the reduced per-word costs with the MT process, our customer saves a staggering amount on translation initiatives each year.
Additionally, certain industries and content types can influence the use of PEMT.
A Common Sense Advisory (CSA) survey to clients showed industry demand for PEMT services is still highly concentrated in IT, heavy equipment and manufacturing. Technical content lends itself to post-editing because traditional MT systems work significantly better on short sentences with limited vocabulary. If you work with terminology-rich, technical content, such as user manuals, service pages, knowledge bases, fact sheets, or FAQs, and if your MT engine has been trained on this specific vocabulary and content, MT results are solid and post-editing works well.
However, thanks to the recent dramatic improvements in MT technology, its appeal is expanding and is expected to increase from sectors that have previously left it alone.
As we stated, the use of PEMT is subjective. We suggest leaning on your translation vendor (*ahem*) as the experts in the industry. They can guide you and offer the tools and resources (such as PEMT) to ensure you can reach or exceed your global business goals.
Making a case for post-edit machine translation (PEMT)
Every day, we create 2.5 quintillion bytes of data — so much that 90 percent of the data in the world today has been created in the last two years alone. And, CSA research shows it takes 14 languages to reach 90 percent of the world’s most economically active populations, but most websites max out with support for just six languages or locales.
Whatever the case may be – increasing content volumes, decreased time, number of languages or stagnant budgets – companies’ ability to stay ahead of the game is being overtaken by the urgent need to keep up, and that need is prevalent. So, what are we really saying here? MT is inevitable. To keep up with the constant rise of content volumes and an increasing “global village,” translations will need to become faster and handle massive content output.
Now, we don’t recommend tossing all of your content through Google Translate and calling it good. But the use of PEMT can be a great solution for your translation needs.
Innovative technology will get us there
We tend to think of ourselves as innovators by profession and localization experts by trade. Not only do we accept change, but we thrive on it. We know technology is here to stay – regardless of the industry – so we’ve embraced it, are learning from it and are working to improve it.
MT technology is advancing all the time. It is becoming more useable, more readable and more customizable every day. Neural MT (NMT) technology, coupled with more consistent, high-quality multilingual resources to feed into the engines, have opened up more areas of application. This is thanks to NMT’s ability to deliver even with less technical, stylistically richer texts. And we are pretty sure we will see ever-higher MT quality before it reaches the post-editors.
Been there, machine translated that
As for our experience, we’ve put together MT solutions for many of the world’s largest companies. We have built statistical and neural engines, and we know how to implement and train third-party MT systems. But what’s more: We combine machine and human capabilities into one unique continuum of services that really matters. Human linguists and MT engines working together to deliver your content across all language borders – at the right time, at the right price, with the right quality.
This article was first published at Acolad. , on March 7, 2023.