If you like the idea of using technology to keep your translation costs to a minimum, you’ll be wanting to look into Machine Translation. If you do, post-editing is something you need to know all about.

Because since the introduction of deep learning to the equation in recent years, MT technology has gotten much better in terms of the quality of output it can produce.

It’s still not suitable for many purposes. Things like marketing, important documents, contracts, e-learning and numerous other creative or high-profile, tone-sensitive projects are still outside MT technology’s grasp.

But for some projects, it’s a great way to give your translation speed a huge boost while minimising costs.

Yet for all those improvements, even the most advanced modern Machine Translation engine still gets things wrong. That’s where post-editing comes in.

The pros and cons of Machine Translation

Machine Translation is a great tool to use for certain types of projects and purposes. These are projects where there tends to be:

  1. A more limited vocabulary – where there is likely to be a limited number of different words and phrases in the content to be translated.
  2. Relatively straightforward sentence structure – complex and compound sentences are going to result in MT output that is much less likely to be accurate.
  3. Few or no mistakes in the original – small spelling errors and typos are easy for a human to spot. An MT engine will be completely stumped by them, however.

This means that any project where there is any nuance or there is likely to be some creativity or cultural understanding required in the translation process will not be suitable for MT.

But in return for only being suitable for projects where the scope is as outlined above, Machine Translation gives you a massive reduction in translation costs over the long term and incredible speed.

Luckily, there is a way to exercise some quality control over Machine Translation output, bringing it up to a standard that’s either suitable for internal company use. Or that’s even comparable to the work of a human translator. That method is called post-editing.

What is post-editing?

Machine Translation post-editing (sometimes abbreviated as PEMT) is the process of correcting errors in content that has been translated by a machine.

The goal in post-editing is to bring the translated content up to an agreed-upon standard. This usually involves doing things like:

  1. Error checking – fixing linguistic and grammatical errors.
  2. Consistency checking – making sure that the structure of the content, spelling, and terminology used are consistent throughout.
  3. Clarity checking – confirming that the content is clear and easy to understand for anyone from the target audience it is intended for.

This sort of thing is necessary for almost all content that has gone through the Machine Translation process. A custom MT engine that has been specifically and extensively trained on the subject matter it’s translating might need very light post-editing, but it will still likely need some.

It’s also important to note that the post-editing process isn’t the same as the kind of editing you get as part of a human-delivered translation service. Human translation editing involves comparing the source and target texts, editing things like style and grammar so that they more accurately reflect the original.

Post-editing of Machine Translation output is much more of a repair process. A post-editor doesn’t even really need to understand the original source language content – they don’t normally consult it.

Ideally, a post-editor needs to be a native speaker of the target language and be a subject matter specialist – understanding how Machine Translation works is also an advantage – but post-editing is much more like having a skilled writer knock the work of a less-skilled writer into shape.

What are the different types of post-editing?

There are two broad types of post-editing. Each is suited to a different sort of project:

1) Light Post-Editing

Sometimes written as LPE, light post-editing is designed to make MT output more readable. There’s no goal of making the output equal to human translation (in fact, this is specified in the ISO 18587 quality standard that relates to it).

If you don’t mind the translation being pretty flat and unnatural, having poor grammar and not much in the way of style, some light post-editing essentially gets the job done.

Of course, it’s easy to think of a hundred projects where this would be completely unsuitable. Yet if you’re looking to translate large volumes of content for internal use or for reference or informational purposes, it’s just the ticket.

Your post-editor will be aiming to use the MT output as much as possible. In the end, your translation might not be very fluent. But it will be accurate and fast to produce.

2) Full Post Editing

Sometimes written as FPE or described as “complete post-editing”, Full Post Editing is what you want when you’re intending to show your MT output to the wider world.

After post-editing, this output should be comparable to that of human translation. This requires much more work on the part of your post-editor. They will be aiming to eliminate all errors and make sure the style, tone, and terminology used are consistent throughout.

Your post-editor may not use all or even that much of the MT output. Unlike in LPE, they will also be comparing the translated target version with your original source version – this requires a post-editor that can speak both of the languages involved.

This all means that FPE is going to take longer than LPE and probably cost more too. But it does mean your Machine Translation project has an output you can use for external purposes and potentially even publication.

Machine Translation, post-editing, and your business

There are a few ways you can effectively build Machine Translation and post-editing into the way your business needs and uses multilingual output:

1) Have a custom MT engine trained by a specialist

There are essentially two different types of MT engines. A generic engine is something like Google Translate. It’s a generalist. Isn’t trained on any one subject area. Thus, the results you get out of this sort of engine tend to be not much more than vaguely legible.

There’s the right time and place for engines like this. Perhaps your budget is effectively zero and you want a rough understanding of a document in another language, for example.

But for most professional tasks, you need a custom MT engine. A custom engine has been trained – most usefully by a Language Service Provider like Asian Absolute that specialises in Machine Translation engine training – on a particular set of data. Bilingual data related to your industry, for example.

This means that the output it produces is likely to be much, much better. Armed with a fully trained custom MT engine, you may only need:

  1. Mono-linguistic expertise – you may only need a post-editor or editors who speak your target language.
  2. Post-editing training – your post-editors will need some understanding of error typologies (the way this engine tends to make mistakes) and details of whether they are aiming for Light Post-Editing or Full Post-Editing.

2) Use an LSP

The simplest MT solution is usually to use an LSP that has established expertise in Machine Translation and the post-editing process. This is particularly the case when you need:

  1. A higher guaranteed quality – when you need the higher-standard Full Post-Editing, for instance.
  2. Short and medium-term cost-effectiveness – looking for a solution that’s more cost-effective than having your own MT engine trained for you, at least in the short and medium-term? An LSP is probably it.
  3. Flexibility – enough to deal with different projects with individual goals and desired outcomes.

An LSP may be able to source or train a custom MT engine trained on in-domain data about your industry. They’ll certainly have access to a solid talent pool of translators with experience in post-editing and be able to provide the level of Quality Assurance you need for each individual project.

What kinds of errors does a post-editor fix?

Thinking about levels of Quality Assurance and error-checking in a project, it’s important to understand the kind of errors in Machine Translation that a post-editor will be aiming to fix – and, perhaps more importantly, those they won’t.

This is something of a tightrope even for very skilled post-editors. Obviously, a skilled linguist like a post-editor has the ability to deliver an extensive re-write of a document that has been through the MT process. But should they?

The overriding goals in most projects that currently involve Machine Translation are speed and cost-effectiveness. A professional translator’s time and skills can be expensive. Minimising their involvement is the whole point of MT in the first place.

A good post editor will iron out the following kinds of errors while holding back from spending too long editing projects where a lighter level of post-editing is called for or the project’s goals don’t call for extreme quality:

  1. Accuracy or cohesion errors – perhaps the engine has translated the same word in different ways in a manner that makes the document inconsistent or confusing?
  2. Grammatical or syntax errors – maybe the MT engine has confused sentence structure or incorporated tags incorrectly?
  3. Readability or fluency errors – perhaps the MT output is too direct of a translation to be correct or read properly in the target language?

Post-editing best practices

1) Set your goals

Whether you’re using your own post-editors and a fully trained custom MT engine that you’ve had developed by an LSP or relying on an LSP to handle the entire Machine Translation process for you, you need to be clear with the people involved what the project’s goals are.

This will allow your post-editors to fine-tune their tightrope-walking to make sure they’re hitting the right balance between quality and speed.

2) Make sure your MT engine is up to the task

If you’re using something not much better than Google Translate to handle the “machine” part of the process, you are going to be making life very difficult for your post-editors.

Conceivably, there may be a situation where this works for what you want from your translated content. But for most purposes, you need a custom MT engine trained on subject matter from your industry.

Otherwise, the post-editing process is likely to be much longer and more costly.

3) Train your post-editors

If you’re using your own fully trained custom MT engine and your own team, make sure you give your post-editors proper training.

How to post-edit should be first, of course. Yet it’s also very helpful for post-editors to understand how Machine Translation works in general and how your specific engine tends to make errors in particular.

4) Use native subject matter specialists

For the fastest, most effective and accurate post-editing, you will want to source post-editors who are subject matter specialists as well as native speakers of your target language who were raised in your target culture.

5) Build in automation

There are at least three tools you can use to give a big boost to the speed and efficiency of the post-editing process:

  1. Involve CAT – using a Computer Assisted Translation tool is a great way to speed up the translation process.
  2. Use TMs – it’s also helpful to have big Translation Memories (abbreviated as TMs, these are databases of pre-translated phrases and terms) that can do some of the work for you.
  3. Multiple-error searches – you might also ensure that your post-editors can easily search for errors that occur multiple times.

Are human translators being replaced?

Machine Translation is nowhere near being able to take over from skilled human translators for any project save for tasks that are high-repetition and high-volume with comparatively low quality requirements.

Even when it comes to Machine Translation, you need to involve expert human translators in the MT engine training process if you want to develop a machine that will create the quality of output you need.

For those projects where it’s a smart choice, Machine Translation will almost always call for some degree of post-editing. Understanding what PE is, how it works, and how you can build it into your company’s processes is important if you want to get the best value from MT as a whole.

Think Machine Translation and post-editing might be the right fit for your next project?

Let’s talk. Asian Absolute specialises in MT engine training and has developed MT solutions for some of the largest organisations in the world.

Set up a cost and commitment-free chat with one of our advisors today. Or request a free, no-obligation quote on demand.