The translation world is changing fast. New technologies appear all the time: neural translation, adaptive translation, rule-based translation, statistical machine translation, adaptive neural translation… it can be hard to keep up. More and more, companies are using these tools in a bid to improve productivity, but the quality of their output still needs to be checked in the old-fashioned, human way. Hence post-editing has become a specialized skill and a crucial step in the translation process. As an English to French translator who loves pretty sentences, I wasn’t sure this would be a task I would enjoy. However, when TAUS asked me to review their post-editing course, I decided it was time to find out more. It is part of the TAUS e-learning platform, which offers courses, but also lots of useful downloadable resources and forums that they’re hoping to expand.
Post-editing and the growth of machine translation
The course starts with a history of machine translation and its context. It goes through different types of machine translation systems: statistical, example-based, rule-based, neural translations and hybrid solutions. The module on evaluation of machine translation systems is very technical again and underlines the need to constantly refine those tools.
The course then moves on to a summary of the post-editing process. It can actually start with pre-editing the source text to ensure its translatability and the efficiency of the process. Ideally, the source text should have limited grammatical structures, simple lexical items and syntactic structures. A style and terminology guide should be used to improve consistency. Whereas translators use subjective, personal methods, post-editing uses specific strategies, based on the type of text, in a bid to improve efficiency: the mistakes that are found in machine translation output are very different from human errors.
This high-level section is very useful to provide background information and gain a general view of translation technology. My understanding of all the different types of tools was rather confused, so this helped me gain a more precise of idea of how each of them works. The modules are packed with information and clearly set out, but there is a lot to learn, so I feel like I may have to go back to it at some point to properly absorb all the new data.
Different types of post-editing
The course then describes the various types of post-editing in greater depth. These depend on the intended use of the final translation and on the quality of machine translation output, which can vary enormously. For all these reasons, post-editing is anything but a “standard” process. For example, the aim of inbound translations is to communicate essential info to non-speakers of the source language. Grammatical and stylistic errors are allowed to stay, and the editing is as light as possible. The raw machine translation output is used as much as possible, and spelling and grammar are secondary to the overall meaning.
On the other hand, full post-editing aims at a result which is similar to human translation, with some compromises in style. The editing mustn’t take longer than a re-translation and a decision to edit or not should be made within two seconds. The aim is to reuse the machine translation output as much as possible and build a better translation around it.
This is a handy dos and don’ts, which gives an idea of what post-editing involves:
- Use as much of the raw MT output as possible.
- Ensure that no information has been added or omitted.
- Basic rules regarding spelling, punctuation and hyphenation apply.
- Ensure that formatting and tags are correct.
- Do NOT pay too much attention to terminology consistency or style.
How to proceed
- Read the machine-translated target segment.
- Make a mental note of the keywords and whether or not the target text seems to make sense.
- Then, read the source segment.
- Compare the meaning of the source and target texts.
- Return to the target text and insert all the necessary changes to make it a grammatically, syntactically and semantically correct reflection of the source text.
The course then describes a good post-editing environment, which can be based around any computer-assisted translation tool, as long as they have an integrated terminology management system and can handle translation memories and export files. It explains how to set up a project in the most productive way possible, and finally goes over feedback, which is essential to retrain the machine translation engine. Indeed, an analysis of the differences between the post-editing output and the raw machine translation data helps to identify error patterns and correct them.
Aiming for a translation which is correct in certain aspects, but deficient in others (mostly stylistically), would be a shock to the system of most translators. We are used to producing the best possible translation in every respect. Leaving an awkward turn of phrase as it is would probably take serious amounts of self-control. This would take practice and discipline, like any new skill.
Post-editing: pricing and required skills
Finally, it ends with the issue of pricing and concludes that there is no standard solution. Indeed, various aspects need to be taken into account, most obviously the quality of the raw machine translation output. Freelancers should give a cost up front with a clear calculation scheme. Again, there is no standard solution and freelancers set their own rates, but they generally amount to 40 to 85% of the translation rate. Freelance post-editors can opt for an hourly rate when the lack of data means that they can’t easily work out the productivity level that can be reached.
To sum up, what skills does a post-editor need? Coud you be a post-editor?
- Excellent command of both languages, of language techonology and of linguistics
- Word-processing skills
- Terminology management
- Solid knowledge of the subject matter
- MT technologies
- Willingness to use new technologies
- Knowledge of different MT systems
- Knowledge of dictionary and terminology management applications
- Ability to adjust expectations and to be pragmatic
After taking this course, I feel like I have a much better understanding of what post-editing is. The course is clearly set out, with attractive graphics and an engaging presentation. I also liked that transcripts were provided, which makes it easier to go back the different sections in search of clarifications.
It also helped me to see that this is very different from translation and even traditional editing. It would require a very different mindset while leveraging my existing skills. Would it bother me to settle for a “good enough” translation? I think that again, it’s all about having clear goals for your work and knowing that some communications don’t need to read perfectly. I think it could be quite interesting to work with clear, objective guidelines in order to produce a text which fulfills a specific role. I'd love to hear from post-editors out there. Do you find the work satisfying? Did you experience a big learning curve? Is it a big part of your activity?
You can also download TAUS PE Guidelines for free!