A hybrid translation approach: Machine Translation Post Editing (MTPE)

MTPE — or machine translation post-editing — has quickly gained prominence as a way to get higher-quality translations at a lower cost than the traditional human-only translation workflow.

Artificial intelligence (AI) has supercharged the traditional MT and MTPE workflows and provided an even better way to get higher-quality translations for less. Smartling customers can leverage Smartling’s AI-powered MT workflow to do most of the heavy lifting. And that means they can reserve their use of human post-editing by expert linguists for their highest-priority translations.

What is machine translation post-editing (MTPE)?

Machine translation post-editing (MTPE) combines the accuracy of human translators with the speed of machine translation. At Smartling, that means a translation process where professional editors review machine-translated output and translation memory matches to ensure high-quality work.

The entire translation process looks like this:

• Machine translation engines translate source text into the target language.
  
• Smartling’s software applies preferred terminology from linguistic assets and edits for spelling or grammar errors.
  
• Professional translators then review the translated text, performing light post-editing as needed to ensure consistency in voice, fluency, and accuracy.

Machine translation may not be a substitute for humans just yet, but it can be a significant first step for companies beginning their localization efforts or expanding into new markets. Suppose you’re looking to ship accurate translations for software or web pages quickly or only need to support one or two languages. In that case, machine translation can be a great choice, especially if you add post-editing services by a professional translator into the mix for an extra layer of quality assurance for translated text.

Is machine translation post-editing right for me?

When considering which translation services to use, it’s important to ask yourself: What quality level is acceptable for the type of content I’m translating? At Smartling, we’ve found that there is an optimal solution for every translation use case, but it may look different for each company.

Let’s get started by comparing MTPE with other popular translation options.

MT vs. MTPE

Neural machine translation offers one of the fastest and most cost-effective ways to translate text into different languages. Most machine translation tools deliver results in minutes or seconds automatically, delivers consistent outputs, and supports a large variety of languages. Finally, as MT has evolved and neural machine translation (NMT) engines have matured, the raw MT output has become more accurate.

Essentially, machine translation engines produces a good base translation. Therefore, MT is an excellent choice for when a base translation is all that’s needed. For example, we often recommend MT-only workflows for content that has a short shelf-life, is time-sensitive, or doesn’t require a strong personality in the translated text. Support tickets, technical manuals, and customer reviews might fall into this category. It’s also an excellent tool for secondary translation efforts like re-translation (editing one or two words rather than submitting an entire paragraph) or pre-translation (getting a sense of how text expansion will impact your UX).

Traditional MTPE supplements the advantages of MT by adding QA done by a professional translator. That allows you to enjoy some time and cost savings while increasing the quality level of the output.

MTPE typically takes two forms, depending on your needs: full post-editing (FPE) or light post-editing (LPE). Light post-editing sweeps the document for glaring issues, like major grammatical errors, spelling mistakes, or incorrect words. That gives you a faster turnaround but isn’t the same depth as FPE, thorough, detailed human editing for style, brand voice, and consistency. Your choice of FPE or LPE depends on the quality of the initial MT output and your use case.

MTPE vs. human translation

Neural machine translation engines can deliver relatively accurate, consistent base translations instantly and for much less money. And professional translators can offer cultural experience and awareness that help improve those translations. So, why not use MTPE for everything?

While it’s true that post-editing can increase the quality of your machine-translated output, it’s not always a replacement for a human translation workflow. Certain content types, such as high-profile marketing content and translations that demand a lot of creativity and cultural sensitivity, require a human touch.

In essence, a professional linguist creates a final product that feels like it was written in your target language with your specific local audience in mind. This isn’t something a piece of software can capture. What’s more, an MT-generated translation doesn’t always provide a great starting point for many of the types of content that you’d want a human-only workflow for.

The more complex, detailed, or sensitive your content is, the more likely you’ll need professional translation services. For that reason, Smartling works with a network of professional translators around the world who specialize not only in making sure your translations are accurate but also that they align with your messaging and sound like they were originally written in that target language.

AI-powered translation workflows

Recent advancements in AI have introduced new ways of working. Those who use AI in translation find they have room in their budgets to translate more content than ever before. They’re also doing so more quickly and without sacrificing quality.

Smartling’s AI Translation workflow uses AI to enhance the traditional machine translation post-editing process. First, we use AI fuzzy match repair to apply your translation memory and repair high fuzzy matches. Then, your strings are sent to multiple MT engines, and AI is used to select the best raw translation. Finally, an AI-powered post-editing step applies preferred terminology from linguistic assets and corrects grammar, syntax, style, and formality.

The translations are delivered to you instantly. Plus, they are of the same quality as you would get from traditional MTPE — for half the cost.

Smartling customers have found that the AI Translation process provides sufficient quality for many use cases where they would have used MTPE before. That includes user guides and agreements, white papers, case studies, and knowledge base and help center articles.

Better yet, they’ve found that this workflow provides an excellent base translation for higher-profile content such as their marketing website and product packaging. Where they once would have used a human-only workflow, they now often engage professional translators and linguists as light post-editors only — and achieve the same level of quality. Human-only translation, therefore, could be reserved for projects that require a very culturally attuned and creative touch.

5 machine translation post-editing best practices

Taking a hybrid approach to your translation needs by blending machine translation with a human post-editing step requires you to follow a few best practices to ensure good results. If you opt to put some of your content through a traditional MTPE, here are a few things to keep in mind.

1. Make sure your source text is as clean as possible

It’s easy to complain about MT accuracy, but often, the culprit isn’t the software; it’s the original text! Awkward sentences, grammatical errors, punctuation mistakes, and inconsistent terminology in the source text can carry over into the machine-translated output.

A lack of consistency can create confusion for your human editors — and, therefore, result in a poor translation. So, it’s important to have someone proofread your source text before translation and make sure it is as accurate and error-free as possible.

This is especially important for more specialized industries like finance or medicine or more elaborate content types like transcription and subtitling.

2. Choose the right TMS for your needs

There are plenty of neural machine translation engines out there, including Google Translate, DeepL, and Microsoft Translator. But what you want is a translation management software (TMS) like Smartling that can manage the entire localization process in one interface — whether you’re ordering machine translation or MTPE or need a more personalized approach through other translation services.

Smartling’s translation management system is a repository for your content and a powerful administrative suite to control everything related to localization:

• See real-time translation progress for your websites, mobile apps, emails, and more in a single snapshot.
  
• Configure Smartling to automatically submit content for translation based on predetermined conditions, like your budget, workflow requirements, and timeline.
  
• Program dozens of quality check options to achieve consistent translations across your content without checking everything yourself.
  
  

The right TMS will also include CAT tools to help the linguists deliver post-editing services, like translation memory, glossary, and style guides for your entire translation team. These tools have been proven to improve accuracy of final translations.

3. Set clear editorial guidelines

Any time you have a team of translators editing your content, you need to have one place to find all the information they need. Document everything in your linguistic asset package. This should include your preferred terminology and your brand’s style (do you use the Oxford comma?) to paragraph length, preferred tone, brand voice, and formality.

That’s because what may be appropriate for your source language doesn’t work for your target language. Think of brands that like to use humor, like Netflix or Taco Bell. They have to carefully calibrate how it will translate text (or not) in other languages because what’s hilarious in English may fall flat in the target language — or worse, offend their audience.

Building a collaborative post-editing process where you work directly with your post-editor sets you up for success (and makes it easier down the line when you’re ready to hire professional translation services).

At Smartling, you’ll get to know your translators via messaging tools like Slack — not only so you can provide feedback more efficiently but also so they get to know your brand and messaging. The more familiar they are with your products and services, the better the edited translations will be.

4. Perfect is the enemy of good

The goal of MTPE is to deliver fast, accurate translations on tight turnaround times — emphasis on fast.

That means automating manual labor as much as possible. Language data network TAUS emphasizes that MTPE should be for a “good enough” approach, which they define as “comprehensible (i.e., you can understand the main content of the message) and accurate (i.e., it communicates the same meaning as the source text).”

In addition to building editorial guidelines, set expectations around what types of errors to edit and how much editors and translators need to touch the translated output. TAUS recommends aiming for semantically correct translations first and foremost and removing any culturally offensive errors, but to use the raw machine translation output as much as possible and to avoid making stylistic changes or editing for sentence flow.

5. Continuously evaluate performance

As you continue to use MTPE, be sure to evaluate your performance. That includes measuring MT quality by assessing how much editing your human linguists do for each project and the overall performance of your pages, including engagement and lead generation.

For those who have or wish to set up LQA workflows for their translations, Smartling offers customers an easy way to manage quality right from within our TMS. While users can perform LQA directly in a translation production project, they can also use our LQA Suite to set up a separate, dedicated project for LQA. Creating samples is simple: users can create “snapshots” of translations from their existing projects. Then, editors can review the translations and log errors. Finally, the edited translations can be seamlessly pushed back to the production environment to ensure accuracy of the final output.

Get MTPE quality for half the cost with Smartling

Traditional MTPE workflows often produce better results than MT-only workflows and are cost-effective compared to traditional human translation workflows. However, there is an even better — and cheaper — way forward. 

Smartling’s AI-powered translation solutions can help you translate more with less. More specifically, with Smartling’s AI translation workflow, companies can get MTPE quality for half the cost. The translations are ready instantly, and quality is guaranteed.

Ready to supercharge your translation efforts with AI and expand your language support? Try Smartling’s AI-powered translation solutions today.

Tags: Machine Translation Human Translation Optimize