What is MTPE?: An Introductory Guide For Businesses

Have you ever needed to get a large load of documents translated, and fast? Let’s say you do, and you call up a translation services company to do the job.

You brace yourself to hear how much it would cost, but you find yourself pleasantly surprised not only at the price in the quote, but also the projected turnaround time. But how is this possible?

This situation is possible if what you’re being offered is machine translation post-editing, or MTPE.

What is Machine Translation Post-Editing?

Some also call it post-edited or post-editing machine translation (PEMT), but machine translation post-editing and MTPE are the more commonly accepted terms. MTPE is a kind of translation process in which a document is first processed through a machine translation (MT) engine, and the output is proofread by a human editor.

One might consider MTPE a hybrid, or even cyborg solution to the problem of high demand in the translation industry. MTPE blends both human skill and machine efficiency to provide a solution that is quick and cost-effective while maintaining a certain standard for the quality of the result.

Types of MTPE

People who are unfamiliar with the field might not know what to expect from MTPE. They might either think that the results of MTPE are inferior to human translation, or they might think that there’s no difference between the two.

Neither side is necessarily wrong. Different kinds of post-editing have different results. Language service providers generally agree that there are two types of post-editing services:

• Light post-editing

• Full post-editing

The primary goal of light post-editing is to make a text understandable. It generally entails proofreading text to remove grammatical errors, and clarifying phrases and sentences that don’t make sense or are completely inaccurate. The quality is usually lower than that of a human translator, but MTPE makes up for this through more affordable costs and quicker delivery. It’s an ideal solution for texts used for internal purposes, or are only necessary for a short time.

Full post-editing is a more thorough approach to the MTPE process, involving not only proofreading and clarification, but also revisions to ensure that the text reads as naturally as a human-translated text. The post-editor works on maintaining a cohesive tone and style, as well as complete accuracy of terms and phrases, during the editing process. In many cases, it also involves taking into account the cultural and linguistic nuances that tend to be lost in machine translation.

Why is post-editing machine translation output feasible?

Machine translation has a history dating back to the invention of the computer, but MTPE is a fairly recent development. In fact, it’s been less than a decade since MTPE was considered a serious option for industrial use!

Much research has gone into the development of machine translation, but for over six decades progress remained below the level where it could actually be useful. But thanks to developments in artificial intelligence and machine learning, MT technology was able to develop to a more sophisticated degree than ever before.

The MT engines of today are capable of providing output whose quality is good enough for human translators to review and make edits more quickly than in the past.

Who can use MTPE and how can they get started?

MTPE isn’t something that only translation services companies can do. Businesses that work with a lot of multilingual data on a regular basis can invest in MTPE capabilities as well.

It isn’t an easy process, however. There is a lot of work to be done in getting the appropriate MTPE solution for any specific business. But here are some of the basics:

Choosing the right MT engine

The first step to investing in MTPE is choosing the right MT engine. There are many different ones out there, and it can be overwhelming to decide which one is the right solution, so here are some points to keep in mind.

The first should be to narrow down the options to MT engines adapted for your specific industry. MT engines that are preset with industry-specific data will provide better results than translation engines trained on linguistic data that is generic.

Another consideration with regard to narrowing down your options is how well the MT engine does in specific language pairs. Be sure to choose an MT engine that performs well in the ones that you need.

Finally—and this is the most important thing—the MT engine that you choose should ideally be one that you can customize with your own data. Even if it’s already adapted to your industry, there may be certain terms or phrasing that are unique to you, or which you’d prefer to leave untranslated. Being able to feed new training data to your MT engine will go a long way toward making it a better fit for your specific needs.

Training/customizing your MT engine

Once you’ve chosen your MT engine, the next step is to customize it. But you might think, today’s MT engines are trained upon billions and billions of translated text segments; would your data even make a dent in that kind of training?

The answer is yes. MT engines don’t rely on quantity alone. The quality of data is also important. Today’s MT engines use AI and deep neural networks to make sophisticated connections that depend on high quality data, which provides necessary context for the MT engine to make the right translation. Sources of high quality data include glossaries, translation memories, and text segments that have already been translated properly.

Results may differ among MT engine providers, and some may require more data than others to come up with a well-trained custom MT engine. But with high quality data, it’s possible to see good results.

Hiring a machine translation post-editor

Once you’re done with your MT engine, it’s time to turn toward finding the right language professional to work with it.

While post-editing work is mostly proofreading and revising MTPE output in one language, it’s not a good idea to give the work to someone who’s monolingual, or unfamiliar with the source language. Machine translation still isn’t perfect, so there may still be errors that only a human translator is able to catch.

You’ll also want your post-editor to have a background in your particular sector. For example, if you’re in law, what you need is a language professional with a legal background. For the healthcare industry, someone also working in medicine would be appropriate. And so on.

Apart from their knowledge of languages in general, knowledge of your industry’s linguistic terms and their contexts will help your post-editor work more effectively.

Pre-editing your source text

Here’s a protip: machine translation works best with simple texts.

It’s always a good idea to look over your source text first to ensure that everything’s grammatically correct and that there are no errors. If you can, take the additional time to change ambiguous phrases, and break down complex sentences into simpler components. Your post-editor will thank you when they receive the better machine-translated output.

Not all translation projects work for MTPE yet

Here’s an important caveat: many documents still need a fully human touch to do properly, often in industries where high creativity or absolute precision is required.

Translating for the legal sector, for example, or for healthcare—these demand no room for errors, and the full scrutiny of a professional translator with a background in those fields. On the other far end is marketing, where the creative element of human language still eludes machine translation.

Still, with continuing advances in MT technology, MTPE, in the form of full post-editing, is starting to become a feasible option in more and more fields.

Conclusion

Machine translation has come a long way to make MTPE a feasible option for businesses. Businesses that want to work with language more efficiently now have many options and the means to get started on their MTPE journey.

Not sure this journey is for your business just yet? Reach out to us at machinetranslation.com, and we’ll handle the rest.