An introduction to post-editing

You may also find it helpful to read our blog article about the challenges of post editing - check it out here.

If you’re new to the online translation world, you might find yourself wondering what post-editing is. If so, that’s great - you’ll be learning an entirely new skill and getting paid to do it!

Simply put, post-editing is the process of converting advanced machine translation into a text that a native speaker could have written. This process takes care of a lot of the issues that we normally encounter with traditional translation - it takes a long time and often strays too far from the source. And this brings us nicely to our first point:

Keeping up with the source!

It’s natural to focus on the target when you’re working on a translation  - after all, that’s the part you can control, and how your success will be measured. But here’s the secret: the source text is actually much more important than the target - that’s where the actual content and meaning comes from. Stray too far from the source, and you’re not actually producing a translated text - you’re writing content.

The great thing is that post-editing keeps you naturally anchored because machine translation is programmed to stay as close to the source as possible. What you need to do now is simple: just go through the text, and in as few edits as possible (and, by extension, as little time as possible) change the text so that it would be indistinguishable from a text written by a native speaker.

What does this mean in practice? Well, let’s say you’re translating from French into English and you’ve got this example. 

Examples 1, 2 and 3 are all semi-acceptable translations, though they can add hidden meanings depending on the context. But here’s the point: the best translation is undoubtedly the one the machine came up with.

You don’t have to prove anything to us by adding in fancy synonyms or embellishing the text - just edit the target until you have a clear cut version of what the source text conveys to the reader.

Work with the machine translation, not against it

You should only remove elements of the machine translation target text where absolutely necessary, usually when the text you have is completely inaccurate. If you can use the structure and vocabulary that are there, please do - excessive edits not only slow you down, making your hourly rate go down, but they also mean that you will most likely stray from the source. 

However, if the machine translation hasn’t worked the way it should, then you will sometimes have to make extensive edits. Being too literal to the point it’s detrimental to the quality of the target text will also result in a poor rating. This might seem like a difficult balancing act when written down like this, but really what we’re saying is: make sure that the translation is accurate and of native quality.

If you delete the machine translation and produce your own version, you will almost certainly fail the training tasks - this is a common error made by professional translators. We’re not asking you to write a new version, we’re asking you to edit the machine translated text - that’s why we call the people who work with us editors.

Check the customer’s instructions and register

Every time you get a new task, it’ll come with instructions from the client and a register: in the vast majority of cases, it’ll be either formal or informal. In limited cases, you’ll see N/A in the register field, which means you should make your own decision based on the source text.

The register isn’t optional, so make sure any edits you make it with the register selected by the customer. Any deviation from what the client has requested will inevitably result in poor evaluation scores.