Machine Translation & Post-Editing: Polished Affordability

It’s no secret that machine translation tools have their limits. Even with the help of artificial intelligence, neural machine translation (NMT) often can’t hold a candle to the results produced by a real live professional translator. In the age of digitalization, however, the amount of text being produced is rapidly increasing – and so is the allure of systems that can translate such content automatically. What can global companies do to satisfy their translation needs without sacrificing comprehensibility?

The answer lies in post-editing. To find out how you can use this process to improve the quality of machine translations and when the combination of MT and post-editing is the right choice, simply read on.

What is machine translation post-editing (MTPE)?

Machine translation uses translation software or an (often free) online translation service to convert text from one language to another. Unfortunately, this fully automated process seldom delivers flawless results. Stilted sentences and improper terminology are quite common, and there’s no way to tailor a text to a particular target audience.

This doesn’t mean you should write off MT tools entirely, though. With a little human assistance, you can turn machine-generated texts into usable content. This is where post-editing by a qualified language expert comes in.

In post-editing, we distinguish between two levels of quality:

1. Light PE: Here, the goal is just to make a text understandable. Style and flow are disregarded entirely, and a certain amount of typos is even tolerated.
2. Full PE: In this case, readers should no longer be able to tell that the text in question was originally produced by a computer. All aspects of style and quality should be on par with translations created by human professionals.

Post-editing can therefore be considered a combination of proofreading and translation.

What advantages does post-editing offer?

Due to the regularity of updates that need to be made available in multiple languages as quickly as possible, software is one prominent area in which machine translation is now indispensable. It would be logistically impossible to have that much text translated by people on such tight deadlines.

The fully automated translation of software documentation – without any human revision whatsoever – is actually now part of the daily routine in the IT industry. In this context, quantity definitely trumps quality.

Meanwhile, machine translation is gaining importance in other sectors, as well. In combination with post-editing, it offers a balance of quality and affordability:

• Simply put, automated translation is usually cheaper than hiring a human translator.
• Post-editing safeguards certain standards of quality.
• An MT system can translate thousands of words in mere seconds, which speeds up the process of bringing products to market.

When does it make sense to post-edit machine translations?

Is post-editing the right option for your texts? When making this decision, you should keep a number of factors in mind.

• Is the text in question meant for internal or external purposes? Translations that only need to provide information to an internal audience are often solid candidates for neural machine translation with light post-editing. If a text is meant for public consumption, however, you should consider full PE – or in some cases, professional translation right from the start (see below).
• What’s your budget? If you’re dealing with large volumes of text, many different languages, and/or a tight schedule, MT might actually be your only viable option. Remember, though, that full post-editing will take a great deal of time and could eat up the savings you were hoping to realize on the translation side.
• How complicated is your source text? Is it clearly written and easy to understand? Is the context readily apparent? Ambiguities and complex structures in your original content can be too much for MT systems to handle. Words with multiple meanings also pose a challenge to these systems, especially when the frame of reference isn’t 100% clear. For example, is the word “table” referring to surface for eating or a chart of figures?
• Does your text contain a lot of repetitions? If phrases, sentences, or even entire paragraphs come up time and again in the same exact form, machine translation can be a sensible option – even in combination with full post-editing. This is particularly true when you have translation memory databases to fall back on.

When should you avoid machine translation?

The tandem of neural MT and post-editing isn’t always enough to ensure that a text will have the desired effect on its target readership. Marketing and advertising content offers a perfect example: In the vast majority of such cases, revising an automated translation doesn’t save much time at all. Going this route can even be more involved – and thus more expensive – than going with a professional from the beginning!

As a general rule, if you want to get an emotional message across, win people over, or maintain your brand’s good image, machine translation systems simply aren’t up to the task.

Furthermore, the use of online tools raises questions related to data privacy law. These free services store the data users enter on the servers of the respective providers, after all, where it’s typically processed further. Of course, confidential information, personal data, and content under copyright should generally not be put online in the first place.

Tips on ensuring quality texts with MTPE

Have you decided to give machine translation and post-editing a shot? The following recommendations can help you increase the quality of automatically translated texts while reducing the overall costs involved.

1. Memento GIGO

The computer scientists out there will already be familiar with the axiom “garbage in, garbage out”, and the same applies to MT: The better your source texts are, the better your results will be when using automated translation.

This is another reason to make sure that your original content is as flawless and comprehensible as possible. More and more technical documentation is being written in plain language, for example, which offers advantages both to users and in the translation process itself. When you invest more time in your texts, it usually won’t take as long to translate and post-edit them.

2. Don’t lose sight of your text’s purpose and audience

With every translation, it’s important to be aware of the intended readers and how you want them to use the content in question.

• Machine translation and light post-editing can suffice for purely informal texts that are meant solely for internal purposes.
• For translations that contain key details or are designed for a more critical audience (or the general public), full PE is essential.

3. Choose your post-editing specialists with care

Post-editing isn’t the same as normal proofreading. Since computers make different mistakes than people do, polishing up machine translations requires particular qualifications and training. The keen-eyed specialists in this area are often translators, but not every translator makes a good post-editor. The requirements a sound post-editing process should meet are defined in ISO Standard 18587:2017.

Our translators are experts in their respective fields, which range from finance and technology to marketing and PR. If you’re playing around with the idea of entrusting your translations to a computer, get in touch with us. We know how machine translation can be a key resource on the path to international success, and we’ll be happy to advise you on the different MTPE options out there. Together, we can fine-tune your texts to your target audience in line with your budget and timeline.