Post-Editing of Machine Translation: Best Practices

10 mins read

Post-Editing of Machine Translation: Best Practices

Each translation and localization project is unique, and so should be your translation approach. Once you decide to implement machine translation (MT), you already guarantee yourself shorter turnaround times and cost reduction. But to maintain the quality of your copy, you’ll need post-editing. This article will focus on what is machine translation post-editing, its types and its benefits.

MTPE or MT Post-Editing Meaning

The term MTPE stands for machine translation post-editing, which is the process of proofreading texts translated by a machine engine. This process aims to improve translations to achieve the same level of output quality as human translation can give.

Machine Translation Post-editing vs Translation

Each of the translation strategies you choose has benefits. You only need to choose which one fits your business approach the most. While the human translation result is a natural, fluent localization that accurately represents the source text, the combination of machine translation and post-editing has other advantages:

Faster Content Turnaround, Even with High-Volume Projects

One of the main benefits of machine translation post-editing is time-saving. A human can translate 3000–3500 words per day based on translators’ answers. Machine translation engines can translate thousands of words in minutes. The time saved by MT is beneficial for projects that involve a high amount of words and a harsh deadline.

Significantly Reduced Costs Compared to Human Translations

Another advantage of machine translation post-editing is cost-saving. With MTPE, a human translator will spend less time translating the text. Thus, the price you pay for localization is mainly for post-editing. This results in a highly cost-effective solution.

Terminology Consistency throughout the Translated Material and High Translation Accuracy

One more benefit of MTPE is that your specific translation memories and glossaries can be integrated and trained to customize and improve the machine engines you use. Suppose you require a regular, not one-time, localization task and have documentation with company or product-specific terms.

Light Editing vs Full Editing

Machine translation also involves choosing the desired level of output quality. Quality is not a clear-cut concept and usually differs from person to person. But generally, the industry defines two levels of post-editing – light and full. Each one brings the raw MT results up to the required level of quality.

Light post-editing involves as few modifications as possible. Its goal is to make the translation understandable, factually accurate, and grammatically correct.

Light post-editing tasks include correcting grammar and spelling mistakes and rewriting confusing sentences, but only if they affect the meaning. Thus, the light post-editing approach covers only critical errors and wrong terminology usage that cause incomprehensible content.

Here you can see the raw MT output and the translation after a light post-editing. In this case, the light post-editing includes 2 changes: adding a preposition ”by” and abbreviation replacement from ”CPC” – machine initials substitution, to ”PDA” – personal digital assistant.

light post-editing example

Full post-editing is a slower and more detail-oriented process. Its goal is to produce accurate translations with consistently correct terminology, appropriate tone, and style. It also includes having no stylistic inconsistencies and grammatical mistakes. After the full post-editing step, the text should look as if a professional linguist worked with it from the very beginning.

Full post-editing tasks include all the light post-editing tasks and cross-referencing translations against other resources, making syntactic modifications in accordance with practices for the target language. It also includes correcting all grammatical errors, typos, punctuation issues, and spelling mistakes. As a result, the content must meet the quality criteria defined by the client for human translations.

Here you can see the raw MT output and the translation after a full post-editing.

full post-editing example

Webinar on how to run multilingual email campaigns with Crowdin and Dyspatch.

Post-Editing Machine Translation: How to Succeed

High-quality MT result comes from the actual engine output and human work. It involves not only the actual post-editing but also reviewing source texts, creating a termbase, and more. Here are a few steps to consider:

Pre-editing for Machine Translation

To ensure the MT and human translator work at their best, you should prepare your source content. The source text should be carefully formatted and pre-edited. Early errors can create problems later on and cause projects delays.

That’s why it’s important to ensure that the original texts have as few errors as possible. As one typo translated into 10 languages is 10 more typos. The main steps to get your content ready for translation include ensuring consistent terminology use and formatting sentences to be straightforward and uncomplicated. For better results, you can edit your copy to be as clear and concise as possible.

Here are some guiding principles for writing or editing your sources:

  • Use an active voice.

  • Beware words with multiple meanings.

  • Use the subject-verb-object sentence structure.

  • Make your sentences short.

  • Use simple grammar.

Machine engines tend to have problems with sarcasm, double-negative, and words with several meanings. Try to avoid them where possible. Learn how to prepare your content for localization from Crowdin.

Localize your product with Crowdin

Customize Your Machine Translation Engine

The scale of the MT market is hard to underestimate. Everyone has what to choose from, depending on the project type and company resources. Some MTs are more suitable for specific language pairs or industries. Choosing the most effective engine can save a lot of time and effort.

On Crowdin, you can localize your website, app, game, landing page, and more with the help of more than 40 machine engines we integrate with.

Along with taking the time to choose the right MT, you can consider customizing the translation engine and training it using your data (glossary, termbase, and previous translations). This will much shorten your way up to high-quality results.

If you want to customize your translation engine, you can install our Crowdin+Intento MT app, which allows you to implement automatic post-editing layers on top of the machine translation and connect your glossaries to some MT engines.

If you are using DeepL Translator, install it from Crowdin Store and customize machine translations with your glossary. Update DeepL MT and try it yourself.

DeepL translator and glossary at Crowdin

Crowdin Quality Assurance Checks

One of the ways you can ensure quality translations and speed up the post-editing step is to enable Crowdin QA checks. This way, translators will dynamically identify any overlooked issues in the translated strings or even new errors that occurred during the post-editing step. Read more about QA checks on Crowdin, how to enable them, what types of errors you can identify, and what Crowdin apps can make this process even more transparent and faster.

Machine Translation Post-editing Must Have: Glossary

Set the terminology base, create and maintain glossaries to help proofreaders ensure consistency. This includes translation memories, glossaries, and any helpful reference or style-guided documents. Remember to keep these files up-to-date for future projects.

MTPE checklist

Post-editing by Human: Who Is the Ideal Candidate

It is important to entrust the post-editing step to a professional, and remember that the post-editor and translator are two separate job positions with different skills and requirements. In 2017, ISO 18587 defined some key aspects of post-editing of machine translation, including the specific skills and competencies of the post-editor.

To receive qualifications and experience, we can recommend visiting webinars, workshops and courses on MTPE. For example, TAUS’ Post-Editing Course and machine translation post-editing courses by DigiLing.

Maximize Quality and Reduce Turnaround Time

Take advantage of machine translation and post-editing services to get the best possible results. If this approach meets your product’s needs, budget and deadlines, you’ll maximize the translation quality and minimize turnaround time.

With the Crowdin platform and the right person to do post-editing, the process would be agile, fast, and clear. Ready to start your next translation project? Let’s get started.

Localize your product with Crowdin

Automate content updates, boost team collaboration, and reach new markets faster.
Diana Voroniak

Previous Post
Node.js i18n: Agile Localization for Developers
Next Post
What’s New at Crowdin: March 2022