Machine Translation (MT) and generative Artificial Intelligence (AI) are changing the multilingual content production game, helping organizations connect with more customers from different linguistic and cultural backgrounds than ever before.
But there’s a dark side to these tools – while they are impressive, they are far from perfect when it comes to accurate translations. It takes human understanding and input to produce quality results, and that’s where post-editing comes in. With MT post-editing (MTPE), a skilled translator edits MT output to correct mistakes and make sure the text reads as it should.
With the rise of advanced Language Learning Models (LLMs) like ChatGPT, we’re seeing different kinds of translation mistakes slip in. LLMs are more creative, sometimes too much so. They can leap to wrong or even offensive conclusions thanks to bias in their training data. They can even “hallucinate”, where translations take a wild turn away from the source content.
In today’s fast-paced world, it’s never been more important for businesses to take advantage of top tools and deploy those tools strategically. In the case of MT, that means creating a comprehensive post-editing strategy to harness the benefits of speed and scale without sacrificing accuracy and quality.
Here are some tips from our experts to help you use post-editing to make sure your AI translations are completed quickly but with quality in mind.
Set clear quality guidelines from the start.
What do we mean by “quality” translation, anyway? You have to answer that question based on your organization’s desired outcomes and use this information to guide your translators through the MTPE process. Your quality target could be different based on the type of material you are translating. User-generated content (UGC) doesn’t need to be top-notch, but legal or compliance content does. When the PEs know what the standards are, you’ll get better results.
Determine when you’ll use light post-editing vs full post-editing.
The two most common post-editing approaches are light post-editing and full post-editing.
Light Post-Editing (LPE): This is the quick-fix approach. Here, the goal is to make minimal changes, just enough to ensure the translation is clear and accurate. Use LPE for projects where speed and budget outrank quality.
Full Post-Editing (FPE): FPE is the more thorough approach, considering not just accuracy but also tone and terminology. It takes longer and costs more, but it guarantees more fluid translations, better cultural relevance, and a higher level of linguistic quality. Choose this approach when you need your translated content to sound as if it were written by a native speaker.
Write with localization in mind.
Clear writing isn't just easier to read—it’s easier to translate, too. Keep your language straightforward, and aim for short, concise sentences.
Avoid idioms, jargon, jokes, cliches, and figures of speech. Finding their equivalents in another language can be tricky even for human translators, and MT has an even more difficult time with them.
Also, choose your words with care. Avoid unnecessary fluff, but don’t skip articles like “the” or prepositions. Humans can often fill in the blanks, but an MT engine will struggle with the ambiguity.
Pre-edit your content.
If you’re not writing content specifically for MT, it’s worth editing it before you send it through the MT engine. Considering the points above, prune and revise your content mercilessly. This might sound like extra work, especially since you’re planning to edit it again afterwards. But your investment will pay off with less time spent post-editing.
If you’ve started using generative AI to create the content, this becomes even more important.
Boost consistency with translation memories and terminology management tools.
Terminology means terms that are specific to your industry. Translators use terminology management tools to make sure that important terms are translated the same way each time. Defining the terms and providing your preferred translation increases consistency, accuracy, and speed.
A translation memory (TM) — a database that contains sentences that have already been translated and allows a translator to automatically plug them in — is another way to boost consistency. Use the language in your translation memory rather than allowing the MT engine to come up with new text. Most translation management systems integrate TM with the MT process.
Use a QA tool to quickly identify errors.
QA tools are like spell-checkers on steroids. You wouldn’t skip running a spelling and grammar check on a Word document, so don’t skip doing it on your MT output either. Like with TMs, most TMS systems have quality checkers integrated.
Train your post-editors.
Your post-editors need more than just linguistic expertise. They also need to understand the quirks of the specific MT model in play. MT engines are not all the same—different models, like NMT and LLMs, can serve up different types of errors.
Train your post editors so they know what to look for. You’ll get better results more quickly.
Define when you don’t really need to post-edit at all
Does MT always need to be edited? Sometimes, the answer is no. However, it’s important to consider this question carefully and pre-define the circumstances under which it’s safe to leave out this step.
You can consider skipping post-editing when the translation is not customer-facing, you need large amounts of content translated as quickly and cheaply as possible, and a rough translation or “gist” is all you need for success. For example, an FAQ or User Generated Content like reviews.
Make MTPE Part of Your Translation Strategy
MT can save you considerable time and money in the translation process, but you need a sound deployment strategy to get the most out of it. MTPE should be part of that strategy. Unnecessary post-editing will add time to your MT workflow and impact your budget. But skimping on it can be an even bigger mistake, since translation errors can damage your brand, diminish your customer experience, or even open your business up to liability.
To strike the balance, your PE program needs to be customized as part of your overall MT deployment strategy. It’s not just about assigning a human to clean up after the machine, you must empower that human with the knowledge, skills, and tools to do the job effectively.
Is your organization ready to take on MT? Take our AI Translation Maturity Assessment to find out. If you’re ready to implement AI translation, our team is here to help set your plan into motion. If you’re not quite there yet, we can help you design a strategy that will guide you to success. Contact us for a consultation today!