12 November 2020
Post-editing machine translations

Post-editing is when a human translator reviews, corrects and improves a machine-generated translation.

It’s an essential process because, while machine translations are fast, they’re usually lacking in quality, accuracy, conviction and style. Not even neural networks can help here. Which means that – at least for the time being – human translators are still at the top of the podium.

What is post-editing?

The best practice for using machine translations has proven to be a two-stage process: Machine Translation Post Editing (MTPE). Once the software has completed the automated translation, a post-editor then makes the necessary corrections. These post-editors are professional translators who are specialised in correcting and optimising machine-generated text.

But just how much manual fine-tuning is required and considered reasonable? That largely depends on how the text will be used.

Purely informative text, for example, contains facts that have to be conveyed correctly and it's vital that the machine translation (MT) doesn’t distort the context. When post-editing this kind of translation, this is the level of intervention required, as stylistics and text flow aren't an issue.

It's a different story when it comes to text with more significance or influence, such as copy used in corporate or commercial publications. Alongside accuracy, the writing style becomes crucial here and the corporate language often needs to be taken into account as well. As a result, sometimes whole sentences or even entire paragraphs have to be rewritten.

So, what are the limitations of MTPE?

The limitations of software-based machine translations and post-editing become abundantly clear when it comes to high-quality technical, legal and specialist text. With the extra time and work involved for the post-edit, one has to wonder if it might be better to hire a human specialist translator from the start.

Even though machine translations have greatly improved through the use of neural software, they're still not perfect. It’s true that grammatical mistakes occur much less frequently these days, but you’d be surprised at how often the meaning is still – quite literally – lost in translation. Unfortunately, these kinds of errors are impossible for a lay person to spot and, without a post-edit, the mistakes get overlooked, leading to potentially serious consequences.

Three typical mistakes made in machine translations

In post-editing, we keep seeing the same flaws and vulnerabilities of machine translations:

Terminology isn’t taken into account: The use of terminology databases ensures that each specific company has a distinctive voice with a consistent, uniform vocabulary. Using an untrained translation system without a database, on the other hand, means that the terminology will have to be reviewed and corrected in the post-edit.

The wrong context: If the context of a text is complicated, a machine translation will often get it wrong and misinterpret the meaning. Post-editing is then necessary to identify these issues and change the wording to reflect the intention of the original text.

Inconsistent use of standard phrases: Poor translations can often be spotted by their obviously incorrect use of expressions and other language conventions. Finding the right level of formality is a classic problem, along with cultural context and proper localisation. Many translation programs have great difficulty at this level, which makes post-editing absolutely essential.

Good quality source text = lower post-editing costs

Google's translation function should really only be used for the most basic purposes. DeepL is somewhat better. However, high-quality MT is only possible with a neural machine translation (NMT) system as used by professional language service providers.

To make sure these systems meet the highest standards possible, they are first trained specifically for the company in question. This process takes a long time and only pays off when the system is used on a regular basis.

Even MT systems trained in this way are not error-free, but the amount of post-editing required is greatly reduced.

What skills does a good post-editor need?

A good post-editor is always a good translator. However, a good translator is not automatically a good post-editor. The demands placed on post-editors are high. So much so, in fact, that these expectations have even been defined in an ISO standard.

Among other things, a good post-editor requires:

  • Translation expertise
  • Linguistic and textual competence in the source and target languages
  • Research skills to obtain and process information
  • Cultural knowledge about the countries where the source and the target languages are spoken
  • Technical competence regarding the function, performance and limitations of machine translation
  • Knowledge of the subject matter being translated

It’s also important that a post-editor doesn’t use their own standards as the benchmark. If, for example, the client has only asked for minimal post-editing for a piece of informative text, then the PE should ignore even minor stylistic flaws. Those who can’t do this, and always approach post-editing with a fine tooth comb, will quickly lose the time gained by the machine translation.

What can you expect from intercontact machine translations?

NMT (neural machine translation) can be the right tool of choice for certain translation tasks. We’ll be happy to advise you about this.

If the conditions are right, we can produce translations faster, thereby saving you around 30% of the costs.

Our professional post-editing service will ensure that you receive a high-quality translation in line with your pre-defined requirements.

Ideally, an MT should never be used without a PE

Despite the hype surrounding machine translation, unedited results should only be used in well-founded, isolated cases. One such example would be with purely informative, straightforward text.

Should you require a higher standard, we always use a combination of machine translation and post-editing. This enables us to save you money AND provide you with high quality in line with your requirements.

However, if your text is important or requires a specialist translation of the highest standard, we still recommend that you use our human translation service.

Talk to us about your specific project and we’ll be happy to advise you about your machine translation and post-editing options.

 Contact intercontact

 


Images sources: