Until just a few years ago, machine translation (MT) was still a niche product. Nevertheless, companies are increasingly incorporating MT into the translation process in order to produce translations in a faster and more cost-effective manner. However, since machine translation still results in errors, post-editing continues to be essential in most scenarios.
In this article, we put our focus on the various types of post-editing and their differences. We explain how to successfully integrate them into the translation process.
An Article by
Flurina Schwendimann
Freelance Translator
It is only possible to get the full added value of machine translation if you combine it with post-editing, which involves revising the results to improve quality.
What exactly is post-editing?
Post-editing involves modifying machine-generated translations with the objective of improving them. Post-editing is necessary to correct the errors in the machine translation while simultaneously reaping the benefits of the price advantages of machine translation. To put it in simple terms, post-editing is a mixture of proofreading and translating, so the term describes the entire process involved in working with machine-produced translations.
Who does post-editing?
Translators and post-editors who specialize in this service and who have undergone appropriate additional training are responsible for post-editing in most cases. The skills that post-editors need to master are enumerated in the relatively new standard ISO 18587. Their qualifications include:
- Translation experience
- Linguistic skill in both the source and target languages
- Cultural and technical expertise
- Knowledge of the corresponding subject area
Typical errors of machine translation systems
The errors that occur in machine translations vary depending on the system. Older rule-based systems generated very choppy translations with poor word order and inappropriate arrangement of clauses. The somewhat more modern statistical systems frequently produced incomplete translations with sentences that were grammatically incorrect. Last but certainly not least, the neural machine translation seen today requires that more attention be paid to content. These systems produce texts of a more stylistically sound and articulate character and have almost no spelling or grammatical errors, making it easier to overlook content errors, especially when they relate to terminology.
Regardless of the system used, it is always advisable to carry out post-editing, with the exception of a few narrow use cases. Depending on the requirements that the target text needs to meet, post-editing may not be absolutely necessary if translations are being produced for test purposes or to be distributed as simple brief posts on social media, for example. A distinction is generally made between light post-editing and full post-editing.
Regardless of the system used, it is always advisable to carry out post-editing, with the exception of a few narrow use cases.
The differences between light and full post-editing
Depending on where the content is being used, the translated text needs to fulfill different criteria. Is the translation only needed in order to provide a general overview (to give a “gist” of the content) or is the translation to be published? Based on the requirements and applicable budget, either light or full post-editing can be carried out. With light post-editing, the post-editor only corrects major errors. With full post-editing, the resulting translation should be (close to) indistinguishable from human translation. TAUS has a series of guidelines for the two different types of post-editing:
Light post-editing
- Delete or supplement information whenever necessary.
- Correct semantic errors in the translation.
- Use the raw translation untouched as much as possible.
- Correct spelling mistakes.
- Revise objectionable or unsuitable content.
- Change the sentence structure only if it impairs comprehension.
- Do not make improvements that are only stylistic in nature.
With light post-editing, only major errors in the machine translation are corrected.
Full post-editing
- Delete or supplement information whenever necessary.
- Revise objectionable or unsuitable content.
- Modify the grammar, syntax, and semantics of sentences as necessary.
- Use corporate terminology.
- Use the raw translation untouched as much as possible.
- Correct spelling, punctuation, and hyphenation errors.
- Adapt the formatting to that of the source text.
After full post-editing, the translation should be indistinguishable from a translation carried out by a human.
Customizable machine translation engine for better raw translations
Keep in mind that the quality of the raw translation can influence the decision regarding what kind of post-editing is ultimately needed. If the raw translation is already quite good, it is sometimes sufficient to carry out light post-editing in order to arrive at a high-quality translation.
If high-quality translations are absolutely required in your area, but you still want to use machine translation, a customized engine can be the right solution. As suggested by the name, these are machine translation systems that have been trained using the customer’s own data, making it possible to deliver better raw translations. You can find more information on this in the article “Machine Translation for Companies".
If the raw translation is already quite good, it is sometimes sufficient to carry out light post-editing in order to arrive at a high-quality translation.
Pre-editing to improve the source texts
The quality of machine translation also depends heavily on the source text. If the source text contains numerous errors or the clauses of the sentences are nested in a convoluted manner, the MT system will have a hard time making sense of it. On the other hand, machine translation will do extremely well with standardized, short sentences containing factual content – such as texts from a business environment, operating instructions and user manuals, reports, and so on.
Pre-editing can therefore be a useful step for increasing the quality of the raw translation and ultimately make post-editing a faster and easier process. You can find more information on this topic in our article “Machine Translation for Translators".
Extra tip
The quality of the source text can be improved through the use of authoring tools. They ensure that certain rules are followed, including ones that improve consistency, bring the writing into line with editorial guidelines, and ensure that corporate terminology is used, resulting in better raw translations and reduced effort in the post-editing stage.
How post-editing can be successfully integrated into the translation process
To include post-editing in the translation process, it is first necessary to implement the machine translation engine. This works best in combination with a translation management system. The desired engine (DeepL, SYSTRAN, KantanMT, Textshuttle, PROMPT, Google Translate, etc.) can be linked with the translation management system through an application programming interface (API). On this basis you create a pre-translation of the text, which is post-edited in the next step.
The clear advantage of this approach is that the post-editor then has access to all relevant content that is available in the translation management system. The key resources are the terminology database, the translation memory, andthe quality management module. Why is all this necessary?
These three TMS components are essential for efficient post-editing, which is known to be all about speed (post-editors process some 7,000 words daily).
- A well-maintained terminology database makes it easier for post-editors to decide which terms to use. Generic engines in particular are prone to errors involving technical terms, which is why the terminology database is worth its weight in gold. For customers, this becomes the basis for fostering the use of corporate terminology.
- A well-populated translation memory ensures that the translations are consistent, with all sentences being stored in both directions so that they can be “recycled” in subsequent projects. If all goes well, this will ensure that post-editors will need to modify fewer segments.
- The quality management module automatically detects errors in the translations and classifies them according to categories. Post-editors have an effective tool for quickly finding and correcting errors.
Post-editing only becomes truly efficient when it is integrated into the systems so that the terminology database, translation memory, and quality management are directly linked.
Machine translation in combination with translation management systems
In theory, companies do not need a translation management system to use machine translation. Texts can be machine translated with the selected engine and then manually sent to the post-editors (such as by e-mail). However, an approach like this has several major disadvantages.
A significant factor here is data security, as internal company information is sometimes processed without any protection. Another pivotal aspect is efficiency during post-editing. If the post-editors lack access to the corresponding technological functions (terminology database, quality management module, or translation memory), post-editing takes longer and requires more effort. This is because the lack of a translation management system means that it cannot be ensured that sentences that have already been translated are translated in the same way again and that the company terminology is adhered to, for example.
Quality management is one of the greatest advantages of combining machine translation and post-editing in the translation management system. Quality criteria are integrated into the system on a centralized basis, allowing you to specifically define such criteria for your individual applications using regular expressions, for example. In this way you can always enable the use of current terminology and automate processes, resulting in faster machine translation, faster post-editing, lower costs, and higher text quality.
You should therefore consider everything very carefully in advance and seek out advice on how you can use the technologies with maximum efficiency and consolidate them if possible.
The bottom line
Machine translation is no longer a technology in its infancy. While it will not replace human translation anytime soon, it is a good alternative that allows you to save time and money in some scenarios. However, translations produced by machines can rarely be used in their “raw” state.
Post-editing is a good solution for taking advantage of machine translation while maintaining a predefined quality standard. Whether light or full post-editing is needed depends on a variety of criteria that the company needs to examine in advance. Machine translation and post-editing have the biggest payoff when used in combination with a translation management system, which provides post-editors with important components that facilitate their work. In addition, it makes it possible for the company commissioning the translation to ensure that data protection requirements are met.
If you are interested in learning more about the topic of machine translation in detail, we recommend our white paper “Introducing Machine Translation” or, if you are looking for more concrete and practical assistance with the process of rolling out machine translation, please check out the new checklist “Achieving success with machine translation”.
Machine translation and post-editing have the biggest payoff when used in combination with a translation management system.
